Risk detection apparatus, risk detection method, and program

Abstract

To provide a risk detection apparatus capable of evaluating the influence in consideration of time lag of an event with respect to business operation.SOLUTION: A keyword extraction unit 101 extracts a keyword related to an event from news data 2. A time-series data generation unit 103 generates score time-series data 6 indicating temporal changes of a risk score that indicates a degree of importance of the event, based on the news data 2 and the keyword. A delay correlation analysis unit 105 calculates delay correlation data 9 based on the score time-series data 6 and operation time-series data 8 indicating temporal changes of an actual result value related to actual results of business operation, the delay correlation data indicating delay correlation which is a correlation in consideration of time lag between the actual result value and the risk score. An output unit outputs notification data based on the delay correlation data 9.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present disclosure relates to a risk detection device, a risk detection method, and a program.

Background Art

[0002] In recent years, the importance of risk assessment and project management has been increasing, and it has been desired to detect events occurring in the external environment and predict the impact of such events on business. However, it is difficult to appropriately extract events that affect business from various events. For example, in the procurement field, it is desired to quickly detect events that pose a risk of supply risk and their impacts, but this is difficult due to the complexity and globalization of the supply chain.

[0003] On the other hand, Patent Document 1 discloses a technique for estimating an abnormality or change in a predetermined event based on event time-series data related to the predetermined event and posting time-series data posted on social media or the like.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] Events occurring in the external environment include not only those that directly and immediately affect business but also those that indirectly affect business over time. That is, there may be a time lag between the occurrence of an event and its impact on business. For example, a cold wave occurring in a certain region induces a large-scale power outage in that region, causing the parts factory existing in that region to stop, which in turn makes it difficult to procure parts over time, resulting in a delay in the delivery date of products.

[0006] In the technology described in Patent Document 1, although it is possible to estimate an abnormality or change in a predetermined event, the time lag of the impact of the abnormality or change in that event on the business is not considered.

[0007] An object of the present disclosure is to provide a risk detection device, a risk detection method, and a program capable of evaluating an impact considering the time lag of an event on a business.

Means for Solving the Problems

[0008] A risk detection device according to an aspect of the present disclosure is a risk detection device that evaluates the impact of an event on a business, and includes an extraction unit that extracts a keyword related to the event from a plurality of article data issued at different dates and times, and based on the article data and the keyword, a generation unit that generates score time-series data indicating a time change of a risk score indicating the importance of the event, and based on business time-series data indicating a time change of an actual value related to the performance of the business and the score time-series data, a calculation unit that calculates delay correlation data indicating a delay correlation relationship, which is a correlation relationship considering the time lag between the actual value and the risk score, and an output unit that outputs notification data based on the delay correlation data.

Effects of the Invention

[0009] According to the present invention, it becomes possible to evaluate an impact considering the time lag of an event on a business.

Brief Description of the Drawings

[0010]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Modes for Carrying Out the Invention

[0011] Hereinafter, embodiments of the present disclosure will be described with reference to the drawings.

[0012] FIG. 1 is a diagram showing a functional configuration of a risk detection device according to an embodiment of the present disclosure. The risk detection device 100 shown in FIG. 1 is a device that evaluates the impact of an event on a business, and includes a keyword extraction unit 101, a geographical feature addition unit 102, a time-series data generation unit 103, a business data aggregation unit 104, a delay correlation analysis unit 105, a risk visualization unit 106, and an impact identification unit 107. The type of business is not particularly limited, and examples include procurement business, safety management business, and management business. The event is, for example, a natural phenomenon, a disaster, an accident related to infrastructure (such as a power outage), and a political event.

[0013] The keyword extraction unit 101 is an extraction unit that uses dictionary data 1 related to a predetermined event to extract keywords related to the event from news data 2, which is text data to be analyzed, and generates keyword data 3.

[0014] The dictionary data 1 includes a risk keyword dictionary indicating keywords related to the event and a thesaurus dictionary indicating the similarity relationship between words. The news data 2 includes a plurality of article data issued at different dates and times. In this embodiment, the article data is news data issued by a predetermined news organization or the like, but may be, for example, posted data posted on social media or the like. The news data 2 is grouped for each unit period (for example, day or week), and the keyword extraction unit 101 generates keyword data 3 for each group.

[0015] The geographical feature addition unit 102 is an addition unit that adds a tag corresponding to the content of the article data to each article data of the news data 2. For example, the geographical feature addition unit 102 adds a tag related to the industry corresponding to the content of the article data to each article data of the news data 2 based on industry geographical data 4 indicating geographical features related to the industry.

[0016] The time series data generation unit 103 is a generation unit that generates score time series data 6 indicating the temporal change of a risk score indicating the importance of an event based on keyword data 3 and geographical characteristic news data 5.

[0017] The business data aggregation unit 104 is an aggregation unit that generates business time series data 8 indicating the temporal change of performance values related to the performance of a business based on business data 7 indicating the performance of the business.

[0018] The delay correlation analysis unit 105 is a calculation unit that calculates delay correlation data 9 indicating a delay correlation relationship, which is a correlation relationship considering the temporal deviation between the risk score and the performance value, based on the score time series data 6 and the business time series data 8.

[0019] The risk visualization unit 106 and the impact identification unit 107 are output units that output notification data including data based on the delay correlation data 9.

[0020] The risk visualization unit 106 is a visualization unit that generates and outputs visualization data obtained by visualizing the risk score based on the keyword data 3 and the score time series data 6. In the present embodiment, the visualization data includes a heat map 10 and a tag cloud 11 for the risk score.

[0021] The impact identification unit 107 is an output unit that generates impact prediction data 12 evaluating the impact of an event on a business based on the delay correlation data 9, and generates and outputs an impact alert report 13, which is notification data for notifying the impact of the event on the business, based on the impact prediction data 12.

[0022] FIG. 2 is a diagram showing an example of the hardware configuration of the risk detection device 100. As shown in FIG. 2, the risk detection device 100 includes an auxiliary storage device 201, a main storage device 202, a processor 203, an input device 204, an output device 205, and a communication device 206. Further, each component (201 to 206) is communicably connected to each other via a bus (not shown).

[0023] The auxiliary storage device 201 is a device that records information and data in a writable and readable manner, and stores a program that defines the operation of the processor 203, as well as various information and data used in that program. The main storage device 202 is used as a work area for processing by the program. The processor 203 reads the program stored in the auxiliary storage device 201 into the main storage device 202 and executes the program using the main storage device 202, thereby realizing the functional units 101 to 107 shown in FIG. 1. The input device 204 is a device that receives various information and data from a user who uses the risk detection device 100. The output device 205 outputs various information to the user, such as by displaying it. The communication device 206 communicates with an external device (not shown) such as a terminal device used by the user, and performs transmission and reception of various information and data with the external device.

[0024] Also, at least a part of the program that defines the operation of the processor 203 may be stored in the recording medium 300. The recording medium 300 stores a computer program and data non-temporarily.

[0025] FIG. 3 is a flowchart for explaining an example of an extraction process for extracting keywords by the keyword extraction unit 101.

[0026] In the extraction process, the keyword extraction unit 101 first executes a dictionary expansion process of expanding the risk keyword dictionary included in the dictionary data 1 based on the synonym dictionary included in the dictionary data 1 (step S101).

[0027] FIG. 4 is a diagram for explaining an example of the dictionary expansion process, and FIG. 5 is a flowchart for explaining an example of the dictionary expansion process.

[0028] As shown in FIG. 4, the dictionary data 1 has a risk keyword dictionary 14 and a thesaurus 15. The risk keyword dictionary 14 includes a field 14A for storing keywords related to events and a field 14B for storing categories of events. The thesaurus 15 includes fields 15A and 15B for storing a plurality of words that are similar expressions to each other, and a field 15C for storing a classification indicating the relationship between these words. The thesaurus 15 may be defined independently of the events. The classification includes "synonym" indicating the same meaning and "similar word" indicating a similar meaning.

[0029] In the dictionary adjustment process, first, the keyword extraction unit 101 starts a loop process B1 that repeats the processes of steps S202 to S205 for each entry in the thesaurus 15.

[0030] In the loop process B1, first, the keyword extraction unit 101 obtains a keyword that matches any of the words included in the target entry from the risk keyword dictionary 14 (step S202). Then, the keyword extraction unit 101 starts a loop process B2 that repeats the process of step S204 for each obtained keyword (step S203). The keyword that matches the word may be, for example, a keyword that exactly matches the word or a keyword that partially matches the word.

[0031] In loop process B2, the keyword extraction unit 101 expands the risk keyword dictionary 14 based on the words included in the target entry and the acquired keywords (step S204). Specifically, when the category included in the target entry is "synonym", the keyword extraction unit 101 adds the word included in the target entry to the risk keyword dictionary 14 as a collation keyword corresponding to the acquired keyword. When the category included in the target entry is "similar word", the keyword extraction unit 101 adds the word included in the target entry as a new keyword and its corresponding collation keyword. When the keyword that matches the word is a keyword that partially matches the word, the keyword extraction unit 101 adds to the risk keyword dictionary 14 a replacement of the matching part in the keyword with the word.

[0032] Then, when the keyword extraction unit 101 executes the process of step S204 for all of the acquired keywords, it ends loop process B2 (step S205). Further, when the keyword extraction unit 101 executes the processes of S202 to S205 for all entries in the similar word dictionary 15, it ends loop process B1 (step S206) and ends the dictionary expansion process. As a result, an extended keyword dictionary 16 in which the risk keyword dictionary 14 is expanded is generated. As shown in FIG. 4, the extended keyword dictionary 16 includes a field 16A that stores a collation keyword, which is a keyword for collation, a field 16B that stores an actually extracted keyword, and a field 16C that stores an event category. Note that the keyword extraction unit 101 adds to the risk keyword dictionary 14 the keywords included in the original risk keyword dictionary 14, associating the keyword itself as a collation keyword.

[0033] Returning to the description of the extraction process. FIG. 6 is a diagram for explaining an example of the processes of steps S102 to S111 of the extraction process. Hereinafter, the extraction process will be described with reference to FIGS. 3 and 6.

[0034] When the dictionary expansion process in step S101 ends, the keyword extraction unit 101 starts a loop process A1 that repeats the processes in steps S103 to S107 for each article data included in the news data 2 (step S102). As shown in FIG. 6, the news data 2 includes a field 2A that stores an article ID for identifying the article data, fields 2B to 2D that store the date, region, and country where the article data was generated, respectively, and a field 2E that stores the text of the article data. Each record of the news data 2 becomes the article data. In this embodiment, the region is a continent, but it is not limited to this example. For example, the region may be the same as the country.

[0035] In the loop process A1, first, the keyword extraction unit 101 extracts characteristic phrases (such as compound words) from the text of the target article data as key phrases, and generates key phrase data 21 indicating the key phrases (step S103). The method for extracting the key phrases is not particularly limited, and examples include methods using algorithms such as TopicRank or MultipartiteRank. The key phrase data 21 includes a field 21A that stores the article ID of the article data from which the key phrase was extracted, a field 21B that stores the key phrase, and a field 21C that stores the number of occurrences of the key phrase in the article data.

[0036] The keyword extraction unit 101 starts a loop process A2 that repeats the processes in steps S105 to S106 for each key phrase included in the key phrase data 21 (step S104).

[0037] In the loop process A2, the keyword extraction unit 101 determines whether there is a match between the target key phrase and a collation keyword included in the extended keyword dictionary 16 (step S105).

[0038] If there is something that matches the collation keyword (step S105: Yes), the keyword extraction unit 101 adds the keyword corresponding to the collation keyword to the extracted keyword data 22 (step S106). The extracted keyword data 22 includes a field 22A for storing the article ID of the article data from which the keyword was extracted, a field 22B for storing the keyword, a field 22C for storing the category of the keyword, and a field 22D for storing the number of occurrences of the keyword in the article data.

[0039] On the other hand, if there is nothing that matches the collation keyword (step S105: No), step S106 is skipped.

[0040] Note that the keyword extraction unit 101 may add a keyword phrase that does not match the collation keyword, together with information indicating that it does not match the collation keyword, to the extracted keyword data 22 as a keyword. Hereinafter, this type of keyword may also be referred to as a non-corresponding keyword. Also, in the example of FIG. 6, by setting the category of the keyword to "non-corresponding", it is shown that the non-corresponding keyword does not match the keyword. The non-corresponding keyword may be used, for example, in the tag cloud 11.

[0041] Then, when the processes of S105 to S106 are executed for all keyword phrases, the keyword extraction unit 101 ends the loop process A2 (step S107). Further, when the processes of steps S103 to S107 are executed for all article data, the keyword extraction unit 101 ends the loop process A1 (step S108).

[0042] Thereafter, the keyword extraction unit 101 starts a loop process A3 that repeats the process of step S110 for each keyword included in the extracted keyword data 22 (step S109).

[0043] In loop process A3, based on the extracted keyword data 22, the keyword extraction unit 101 calculates a score indicating the importance of each extracted keyword for each keyword, and generates keyword data 3 indicating each keyword and its score (step S110). The score is calculated based on the number of occurrences, which is the number of times the keyword appears in the news data 2. Here, it is calculated so that the importance of a keyword that appears frequently in a specific article data but infrequently in the entire article data becomes high. This type of score is calculated, for example, by using algorithms such as TF-IDF (Term Frequency - Inverse Document Frequency) and its subtypes (such as sublinearTF). The keyword data 3 includes a field 3A for storing the article ID of the article data from which the keyword was extracted, a field 3B for storing the keyword, a field 3C for storing the category of the keyword, and a field 3D for storing the score of the keyword.

[0044] When the processing of step S110 is completed for all keywords, the keyword extraction unit 101 ends loop process A3 (step S111), and then ends the extraction process.

[0045] The extraction process described above is merely an example and is not limited thereto. For example, the keyword extraction unit 101 may extract keywords from the text of the article data using the extended keyword dictionary 16 without performing the process of extracting key phrases in step S103. Also, the keyword extraction unit 101 may extract keywords using the original risk keyword dictionary 14 without creating the extended keyword dictionary 16.

[0046] FIG. 7 is a flowchart for explaining an example of the addition process of adding tags related to industries by the geographical feature addition unit 102, and FIG. 8 is a diagram for explaining an example of the addition process.

[0047] In the addition process, first, for each article data included in the news data 2, the geographical feature addition unit 102 starts a loop process C1 that repeats the processes of steps S302 to S309 (step S301).

[0048] In the loop process C1, first, the geographical feature addition unit 102 extracts a place name keyword, which is a keyword related to a place name, from the text of the target article data, and generates place name keyword data 31 indicating the place name keyword (step S302). The method for extracting the place name keyword is not particularly limited. For example, a method of applying a named entity recognition (NER) algorithm to the text of the target article data and extracting a named entity classified as a place name as the place name keyword can be mentioned. The place name keyword data 31 has a field 31A for storing the article ID of the article data from which the place name keyword is extracted, and a field 31B for storing the place name keyword.

[0049] For each place name keyword included in the place name keyword data 31, the geographical feature addition unit 102 starts a loop process C2 that repeats the processes of steps S304 to S308 (step S303).

[0050] In loop process C2, first, the geographical feature addition unit 102 converts the target place name keyword into place name information indicating a geographical location, and generates place name data 32 including the place name information (step S304). Note that the place name keyword extracted by an algorithm for extracting proper expressions does not indicate a geographical location but simply represents a place name, so this process is performed. Also, there are variations in the notation of the place name keyword. The geographical feature addition unit 102 generates the place name data 32, for example, by collating a previously prepared city database with the place name keyword. At that time, it is preferable for the geographical feature addition unit 102 to further perform a process of standardizing the notation. Examples of this process include adding or deleting a suffix such as "State", and regularizing abbreviations (such as "TX (Texas)") and abbreviated forms (such as "St (Saint)"). The place name data 32 has a field 32A for storing the article ID of the article data from which the place name keyword converted into place name information is extracted, and fields 32B and 32C for storing the place name information. In this embodiment, the place name information includes the country name and the state name or prefecture name, field 32B stores the country name, and field 32C stores the state name or prefecture name. However, the place name information is not limited to this example, and for example, another regional name may be used instead of the state name or prefecture name, and the unit of the regional name may differ for each country.

[0051] Subsequently, the geographical feature addition unit 102 specifies an industry related to the geographical location indicated by the place name information based on the industrial geographical data 4 (step S305).

[0052] FIG. 9 is a diagram for explaining an example of industrial geographic data 4 and a method for generating the same. The industrial geographic data 4 shown in FIG. 9 has a field 4A for storing country names, a field 4B for storing state names or prefecture names, and a field 4C for storing industries related to geographical locations indicated by country names and state names or prefecture names. The industries related to geographical locations are, for example, industries that are prosperous in those geographical locations, and are defined by products produced in factories provided in those locations, etc. The industrial geographic data 4 is generated, for example, from an enterprise database 41 or industrial statistical information 42. The enterprise database 41 is a database that comprehensively collects information related to enterprises, and includes, for example, enterprise names, industries, country names, locations, and sales amounts. The industrial statistical information 42 is statistical information related to industries, and includes country names, place names, industries, and sales amounts, etc. The industrial geographic data 4 is generated, for example, by aggregating sales amounts by industry for each geographical location based on the enterprise database 41 or the industrial statistical information 42, and selecting industries with sales amounts equal to or higher than a reference value at each location. The reference value is, for example, a predetermined ratio (e.g., 1%, etc.) of the total sales amount in the industry. Also, the industrial geographic data 4 may be generated by the risk detection device 100, or may be generated by other devices.

[0053] Return to the description of the additional process. The geographical feature addition unit 102 starts a loop process C3 that repeats the process of step S307 for each identified industry (step S306).

[0054] In the loop process C3, the geographical feature addition unit 102 generates tag data 36 in which the target article data (article data from which a place name keyword converted into a place name corresponding to the target industry is extracted) is associated with the target industry as a tag, and generates data including the news data 2 and the tag data 36 as geographical feature news data 5 (step S307). The tag data 36 includes a field 36A for storing an article ID for identifying the target article data, a field 36B for storing a tag type which is the type of the tag, and a field 36C for storing the tag. The tag type is, here, "industry".

[0055] Then, when the geographical feature addition unit 102 finishes the process of step S307 for all industries, it ends loop process C3 (step S308). Also, when the geographical feature addition unit 102 finishes the processes of steps S304 to S308 for all place name keywords, it ends loop process C2 (step S309). Further, when the geographical feature addition unit 102 finishes the processes of steps S302 to S309 for all article data, it ends loop process C2 (step S310) and ends the addition process.

[0056] The addition process described above is merely an example and is not limited thereto. For example, if a city name or the like is added to the article data, the geographical feature addition unit 102 may use the city name or the like as a place name keyword without performing named entity extraction. Also, if coordinate information (for example, latitude and longitude) is added to the article data, the geographical feature addition unit 102 may extract, as a place name keyword, a city near the position indicated by the coordinate information based on the coordinate information and a city database or the like.

[0057] Also, the geographical feature addition unit 102 may generate tag data 36 in which tags other than the tag type of "industry" are associated with the article data by using a keyword mapping method. FIG. 10 is a diagram for explaining an example of a method for generating tag data 36 using a keyword mapping method.

[0058] As shown in FIG. 10, a tagging target list 37 is prepared in advance. The tagging target list 37 includes a field 37A for storing the category of the event related to the keyword and a field 37B for storing the keyword. In the field 37B, as shown as the tagging target list 37-1, individual keywords may be stored, or as shown as the tagging target list 37-2, an all-designation symbol (in the example of the figure, "*") for designating all the keywords included in the category may be stored. When there is a combination in the tagging target list 37 that matches the combination of the keyword and the category in the keyword data 3, the geographical feature addition unit 102 adds the category of that combination as the tag type and the keyword of that combination as the tag to the tag data 36. In the example of FIG. 10, "material" is shown as the category, but it is not limited to this.

[0059] FIG. 11 is a flowchart for explaining an example of a first generation process for generating score time series data 6 from keyword data 3, and FIG. 12 is a diagram for explaining an example of the first generation process.

[0060] In the first generation process, the time series data generation unit 103 executes a loop process D1 that repeats step S402 for each keyword in the keyword data 3 (step S401).

[0061] In the loop process D1, the time series data generation unit 103 aggregates the score of the target keyword for each combination of the date and region corresponding to the article ID of the article data from which the target keyword is extracted based on the news data 2 and the keyword data 3 to generate first aggregated data 41 (step S402). Here, the aggregation unit of the date is "day" and the region is the continent. As shown in FIG. 12, the first aggregated data 41 includes a field 41A for storing the date when the article data was generated, a field 41B for storing the continent where the article data was generated, a field 41C for storing the keyword extracted from the article data, and a field 41D for storing the score aggregation value obtained by aggregating the scores of the keywords.

[0062] When the time series data generation unit 103 finishes executing the process of step S402 for all keywords, it ends the loop process D1 (step S403).

[0063] Then, based on the first aggregated data 41, the time series data generation unit 103 generates the first score time series data 61, which is used as the score time series data 6, by calculating, for each combination of date and region, the total value of the respective aggregated values of the keywords corresponding to that combination as the region score (step S404), and then ends the process. As shown in FIG. 12, the first score time series data 61 includes a field 61A for storing the date on which the article data was generated, a field 61B for storing the region where the article data was generated, and a field 61D for storing the region score.

[0064] FIG. 13 is a flowchart for explaining an example of a second generation process of generating the score time series data 6 from the geographical feature news data 5 by the time series data generation unit 103, and FIG. 14 is a diagram for explaining an example of the second generation process.

[0065] In the second generation process, the time series data generation unit 103 executes a loop process D2 that repeats step S412 for each keyword in the geographical feature news data 5 (step S411).

[0066] In the loop process D2, based on the keyword data 3 and the geographical feature news data 5, the time-series data generation unit 103 aggregates the scores of the target keywords for each combination of the date corresponding to the article ID of the article data from which the target keywords are extracted and a predetermined perspective, and generates second aggregated data 42 (step S412). Here, the aggregation unit of the date is "week". As shown in FIG. 14, the second aggregated data 42 includes a field 42A for storing the date when the article data was generated, a field 42B for storing the perspective type which is the type of the aggregated perspective, a field 42C for storing the aggregated perspective, a field 42D for storing the keyword, and a field 42E for storing the score aggregation value obtained by aggregating the scores of the keywords. The perspectives are, for example, countries, industries and materials which are tags, etc.

[0067] When the time-series data generation unit 103 executes the process of step S412 for all the keywords, it ends the loop process D2 (step S413).

[0068] Then, based on the second aggregated data 42, the time-series data generation unit 103 generates second score time-series data 62 in which the total value of the respective score aggregation values of the keywords corresponding to the combination is calculated as the perspective score for each combination of the date and the perspective, and sets it as the score time-series data 6 (step S414), and ends the process. As shown in FIG. 14, the second score time-series data 62 includes a field 62A for storing the date when the article data was generated, a field 62B for storing the perspective type which is the type of the perspective, a field 62C for storing the perspective, and a field 62D for storing the perspective score.

[0069] Note that the time-series data generation unit 103 may convert the second score time-series data 62 into second score time-series data 63 which is data associating the date with the perspective score for each perspective. As shown in FIG. 14, the second score time-series data 63 includes a field 63A for storing the date and a field 63B for storing the perspective score. In the example of FIG. 14, as the perspective score, an American risk score where the perspective type is a country and the perspective is the perspective score of the United States, and a petrochemical risk score where the perspective type is an industry and the perspective is petrochemical are shown.

[0070] The first score time-series data 61 and the second score time-series data 62 (the second score time-series data 63) generated by the first generation process and the second generation process described above are examples of the score time-series data 6. Also, the regional score and the perspective score are examples of risk scores indicating the importance of events. Note that the regional score can also be regarded as a perspective score from the perspective of a region. The first score time-series data 61 may be converted into the same format as the second score time-series data 63.

[0071] FIG. 15 is a diagram for explaining an example of a business generation process for generating business time-series data 8 from business data 7 by the business data aggregation unit 104.

[0072] In the present embodiment, the business is the purchase of goods, and the business data 7 is purchase performance data of purchased goods. In this case, as shown in FIG. 15 for example, the business data 7 includes a field 7A for storing the name of the purchased goods, a field 7B for storing the goods category which is the category of the name, a field 7C for storing the order date of the goods, and a field 7D for storing the arrival date when the goods arrived.

[0073] First, based on the business data 7, the business data aggregation department 104 calculates, for each business, a delay index, which is an index indicating the delay that occurred in the business, as a business index related to the business, and generates the indexed business data 71 with the delay index added to the business data 7. In this embodiment, the delay index is the lead time (LT: Lead time), which is the number of days from the order date to the arrival date. However, the delay index is not limited to the lead time, and may be an index using the order date, arrival date, promised delivery date, and values related thereto, such as the number of delayed days, which is the difference between the promised delivery date and the arrival date, the difference between a predetermined standard lead time and the lead time, or the number of adjusted cases obtained by adjusting the promised delivery date. Also, the business index is not limited to the delay index, and may be, for example, the price of the item or the difference between the price of the item and the standard price. As shown in FIG. 15, the indexed business data 71 further includes a field 7E for storing the lead time in addition to the same fields 7A to 7D as the business data 7.

[0074] Subsequently, the business data aggregation department 104 calculates an aggregated delay index value obtained by aggregating the lead time in the indexed business data 71 for each combination of date and item category, and generates aggregated business data 72 indicating the aggregated delay index value. The aggregated delay index value is an example of an actual value related to the performance of the business. In this embodiment, it is the LT median, which is the median of the lead time, which is the delay index. However, the actual value is not limited to this example, and may be the sum, average value, and maximum value of the lead time. The aggregation unit of the date is the same as the aggregation unit of the date of the article data (here, "week"). The aggregated business data 72 includes a field 72A for storing the date, a field 72B for storing the item category, and a field 72C for storing the LT median, which is the median of the lead time corresponding to the combination of the date and the item category. Note that the collection unit is not limited to the item category, and may be the manufacturer of the item, a predetermined term in the product name of the item, or the like.

[0075] Then, the business data aggregation unit 104 generates the business time-series data 8 by converting the aggregated business data 72 into data indicating the LT median value of each item category for each date. As shown in FIG. 15, the business time-series data 8 includes a field 8A for storing the date and a plurality of fields 8B for storing the LT median values of the respective categories.

[0076] FIG. 16 is a flowchart for explaining an example of an analysis process for generating the delay correlation data 9 by the delay correlation analysis unit 105.

[0077] In the analysis process, the delay correlation analysis unit 105 starts a loop process E1 that repeats the processes of steps S502 to S506 for each explanatory variable, with each perspective score of the score time-series data 6 as an explanatory variable (step S501).

[0078] In the loop process E1, the delay correlation analysis unit 105 starts a loop process E2 that repeats the processes of steps S503 to S505 for each target variable, with each LT median value of the business time-series data 8 as a target variable (step S502).

[0079] In the loop process E2, the delay correlation analysis unit 105 executes a delay correlation process for calculating delay correlation information indicating a correlation relationship considering the time difference between the target explanatory variable and the target target variable (step S503).

[0080] FIG. 17 is a diagram for explaining an example of the delay correlation process. In FIG. 17, the target explanatory variable is the j-th explanatory variable e j , and the target target variable is the m-th target variable o m . Also, the cumulative period for accumulating the explanatory variable is the addition coefficient (i - j), the change period during which the target variable changes is the difference coefficient (m - n), and the completion period until the change of the target variable is completed is the delay coefficient (m - i).

[0081] The delay correlation analysis unit 105 accumulates the explanatory variable according to the addition coefficient (i - j) to obtain the cumulative explanatory variable e ij (= e j + e j+1 + … + ei ) and a differential target variable O obtained by calculating a difference according to a delay coefficient (m - i) and a difference coefficient (m - n) with respect to the target variable mn (= e m - e n ) is calculated by comprehensively changing each coefficient (addition coefficient (i - j), difference coefficient (m - n), and delay coefficient (m - i)), and the largest correlation coefficient among the calculated correlation coefficients is calculated as the delay correlation coefficient. Here, calculating the correlation function by comprehensively changing each coefficient specifically means calculating the correlation function for all combinations of coefficients included in the predetermined coefficient ranges for each coefficient.

[0082] The delay correlation information includes the above - mentioned delay correlation coefficient and each coefficient (addition coefficient (i - j), difference coefficient (m - n), and delay coefficient (m - i)) used for calculating the delay correlation coefficient. The delay correlation coefficient is a coefficient representing the strength of the correlation between the explanatory variable and the target variable, and the delay coefficient is a coefficient representing the magnitude of the time lag at which the correlation between the explanatory variable and the target variable occurs.

[0083] Returning to the description of the analysis process. The delay correlation analysis unit 105 determines whether the calculated delay correlation coefficient is significant (step S504). Here, the delay correlation analysis unit 105 determines whether the delay correlation coefficient is greater than or equal to a predetermined value. If the delay correlation coefficient is greater than or equal to the predetermined value, it is determined that the delay correlation coefficient is significant. If the delay correlation coefficient is less than the predetermined value, it is determined that the delay correlation coefficient is not significant.

[0084] When the delay correlation coefficient is significant, the delay correlation analysis unit 105 adds the delay correlation coefficient and coefficient - related information, which is information related to the delay correlation coefficient, to the delay correlation data 9 (step S505). On the other hand, when the delay correlation coefficient is not significant, the delay correlation analysis unit 105 skips the process of step S505. The coefficient - related information includes the target variable and each coefficient (addition coefficient, difference coefficient, and delay coefficient).

[0085] Then, when the delay correlation analysis unit 105 executes the processes of steps S503 to S505 for all target variables, it ends the loop process E2 (step S506). Also, when the delay correlation analysis unit 105 executes the processes of steps S502 to S506 for all explanatory variables, it ends the loop process E1 (step S507) and ends the process.

[0086] FIG. 18 is a diagram showing an example of the delay correlation data 9. The delay correlation data 9 shown in FIG. 18 includes fields 9A to 9E. The field 9A stores the explanatory variable, and the field 9B stores the target variable. The field 9C stores the delay correlation coefficient between the explanatory variable and the target variable, and the fields 9D to 9F store the addition coefficient, the delay coefficient, and the difference coefficient used for calculating the delay correlation coefficient, respectively.

[0087] The analysis process described above is merely an example and is not limited thereto. For example, the addition coefficient and the difference coefficient may be fixed values. For example, the addition coefficient may be 1 and the difference coefficient may be 0. Also, in the delay correlation process, the difference of the explanatory variable may be calculated according to the difference coefficient, and the target variable may be accumulated according to the addition coefficient.

[0088] As described above, the risk visualization unit 106 generates and outputs the heat map 10 and the tag cloud 11 as visualization data for visualizing the influence of events on business based on the keyword data 3 and the score time series data 6.

[0089] FIG. 19 is a diagram showing an example of the heat map 10. The heat map 10 shown in FIG. 19 is a matrix-type graph with the horizontal axis as the time axis (date) and the vertical axis as the region, and each value shows a color reflecting the risk score corresponding to the date and the region. The color reflecting the risk score is, for example, gray when the risk score is low, and the redder it becomes as the risk score increases. Note that the vertical axis is not limited to the region and may be from other viewpoints or may be specified by the user.

[0090] FIG. 20 is a diagram showing an example of the tag cloud 11. The tag cloud 11 shown in FIG. 20 is, for example, a graph showing each keyword extracted from the article data corresponding to the date and region of the value when the value of the heat map 10 is selected. The display method of the keywords in the tag cloud 11 is not particularly limited. For example, in the tag cloud 11, it may be possible to display or not display place names, and select the category of keywords to be displayed. Also, the tag cloud 11 may show the keywords in sizes and colors corresponding to their risk scores. Also, the place names and other keywords may be shown in different colors. When a keyword shown in the tag cloud 11 is selected by the user, the risk visualization unit 106 may output the text of the article data from which the keyword was extracted. At that time, the risk visualization unit 106 may emphasize the selected keyword in the text of the article data, such as by highlighting it.

[0091] Note that the visualization data is not limited to the heat map 10 and the tag cloud 11. For example, the visualization data may be map information that shows the region of the value selected in the heat map 10 on a map such as a world map.

[0092] FIG. 21 is a flowchart for explaining an example of the influence identification process by the influence identification unit 107.

[0093] In the influence identification process, the influence identification unit 107 acquires all the explanatory variables having values (risk scores) equal to or higher than a risk threshold, which is a predetermined threshold, from the lag correlation data 9 as risk explanatory variables (step S601).

[0094] The influence identification unit 107 starts a loop process F1 that repeats the processes of steps S603 to S607 for each of the acquired risk explanatory variables (step S602).

[0095] In the loop process F1, the influence identification unit 107 extracts the lag correlation coefficient of the target risk explanatory variable from the lag correlation data 9 (step S603).

[0096] The influence identification unit 107 starts a loop process F2 that repeats the process of step S605 for each extracted delay correlation coefficient, which is the extracted delay correlation coefficient (step S604).

[0097] In the loop process F2, the influence identification unit 107 acquires coefficient-related information, which is information related to the target extracted delay correlation coefficient, and adds the acquired explanatory variable, the extracted delay correlation coefficient, and the coefficient-related information to the influence prediction data 12 (step S605).

[0098] Then, when the influence identification unit 107 executes the process of step S604 for all the extracted delay correlation coefficients, the loop process F2 ends (step S606). Further, when the influence identification unit 107 executes the processes of steps S603 to S606 for all the risk explanatory variables, the loop process F1 ends (step S607).

[0099] The influence identification unit 107 generates and outputs an influence alert report 13 based on the influence prediction data 12 (step S608), and ends the process.

[0100] FIGs. 22 and 23 are diagrams showing an example of the influence alert report 13. The influence alert report 13 shown in FIG. 22 includes a setting unit 131, a selection unit 132, a score display unit 133, and a coefficient display unit 134.

[0101] The setting unit 131 is an interface for setting a risk threshold. The setting unit 131 may visualize the range of the explanatory variables to be alerted by the set risk threshold by displaying a histogram of the risk scores.

[0102] The selection unit 132 is an interface for selecting the type of perspective for visualizing the impact on the business. The type of perspective to be selected may be single or multiple. All types of perspectives may be set as default for visualization. Also, the selection unit 132 may highlight the type of perspective of the perspective having a risk score equal to or higher than the risk threshold. In the example of the figure, the highlighting is realized by assigning a predetermined mark 135, but it may be realized by highlighting the characters indicating the type of perspective or the like.

[0103] The score display unit 133 displays the perspectives included in the type of perspective selected by the selection unit 132 and their risk scores. Also, the score display unit 133 may display only the risk score of the perspective selected by the user among the perspectives included in the type of perspective selected by the selection unit 132. Also, the score display unit 133 may highlight the perspectives having a risk score equal to or higher than the risk threshold. The method of highlighting is the same as the highlighting by the selection unit 132.

[0104] The coefficient display unit 134 displays the delay correlation coefficient and coefficient-related information between the category of the business (item) correlated with the perspective included in the type of perspective selected by the selection unit 132 and the LT median value which is the actual value of the category. The coefficient-related information displayed in the figure is only the delay coefficient, but the addition coefficient and the difference coefficient may be displayed.

[0105] The impact alert report 13 shown in FIG. 23 includes a score graph 141, a tag cloud 142, and map information 143.

[0106] The score graph 141 is a graph of the score time-series data 6 showing the change in the risk score of the perspective selected by the score display unit 133 in FIG. 22. A line indicating the risk threshold may be further shown in the score graph 141.

[0107] The tag cloud 142 is a graph showing each keyword used in calculating the risk score of the date selected in the score graph 141 from the perspective selected in the score display section 133 of FIG. 22. The method of displaying keywords in the tag cloud 142 may be the same as that of the tag cloud 11.

[0108] When the perspective selected in the score display section 133 of FIG. 22 is a country or region, the map information 143 represents that country or region on the map.

[0109] As described above, according to the present embodiment, the keyword extraction unit 101 extracts keywords related to an event from the news data 2. The time-series data generation unit 103 generates score time-series data 6 indicating the time change of the risk score indicating the importance of the event based on the news data 2 and the keywords. The delay correlation analysis unit 105 calculates delay correlation data 9 indicating a delay correlation relationship, which is a correlation relationship considering the time lag between the performance value and the risk score, based on the business time-series data 8 indicating the time change of the performance value related to the business performance and the score time-series data 6. The output unit (risk visualization unit 106 and impact identification unit 107) outputs notification data based on the delay correlation data 9. Therefore, since the delay correlation data 9 indicating the delay correlation relationship, which is a correlation relationship considering the time lag between the performance value related to the business performance and the risk score indicating the importance of the event, is calculated, it is possible to evaluate the impact considering the time lag of the event on the business.

[0110] Also, in the present embodiment, the time-series data generation unit 103 generates the score time-series data 6 based on the number of appearances, which is the number of keywords appearing in the article data. In this case, it is possible to more appropriately reflect the importance of the event in the score time-series data 6.

[0111] Furthermore, in this embodiment, the time-series data generation unit 103 generates score time-series data 6 based on the aggregated value of the number of occurrences of keywords related to each perspective regarding the event. The delayed correlation analysis unit 105 calculates delayed correlation data 9 for each perspective. This makes it possible to evaluate the impact of an event on business operations from multiple perspectives, taking into account time lags.

[0112] In this embodiment, the time-series data generator 103 generates score time-series data 6 corresponding to each viewpoint for each viewpoint type. The viewpoint type includes, for example, at least one of region, country, industry, and material. In this case, it becomes possible to evaluate the impact of an event on business operations from multiple angles, taking into account time lags.

[0113] Furthermore, in this embodiment, the geographical characteristic adding unit 102 adds industry-related tags to each piece of article data included in the news data 2, based on the geographical location related to the content of the article data and the industrial geographical data 4 indicating geographical characteristics related to the industry. The time-series data generating unit 103 generates score time-series data 6 from an industry perspective based on the tags. In this case, it becomes possible to evaluate the impact of an event on business operations for each industry, taking into account the time lag.

[0114] In this embodiment, the geographical characteristic adding unit 102 adds a category-related tag to each piece of article data based on the category of the keyword extracted from the article data. The time-series data generating unit 103 generates score time-series data 6 based on the tag, with the category being the perspective. In this case, it is possible to generate score time-series data 6 with various categories being the perspective depending on the type of keyword, etc.

[0115] In this embodiment, the delayed correlation data 9 includes a correlation coefficient (delayed correlation coefficient) that indicates the strength of the correlation between the performance value and the risk score, and a delay coefficient that indicates the magnitude of the time lag at which the correlation occurs between the performance value and the risk score. In this case, it becomes possible to more appropriately evaluate the impact of an event on business operations, taking into account the time lag.

[0116] Also, in the present embodiment, the delay correlation analysis unit 105 calculates a correlation coefficient and a delay coefficient of a viewpoint whose delay correlation coefficient is equal to or greater than a predetermined value as delay correlation data 9. In this case, it becomes possible to more appropriately evaluate the influence on the business.

[0117] Also, in the present embodiment, the notification data includes information indicating a risk viewpoint, which is a viewpoint in which a risk score equal to or higher than a threshold value exists, among the viewpoints corresponding to the delay correlation data 9. The notification data also includes a heat map and a tag cloud. In these cases, it is possible to more appropriately notify the influence on the business.

[0118] The above-described embodiments of the present disclosure are examples for explaining the present disclosure, and are not intended to limit the scope of the present disclosure only to those embodiments. Those skilled in the art can implement the present disclosure in various other modes without departing from the scope of the present disclosure.

Explanation of Reference Numerals

[0119] 100: Risk detection device 101: Keyword extraction unit 102: Geographical characterization unit 103: Time-series data generation unit 104: Business data aggregation unit 105: Delay correlation analysis unit 106: Risk visualization unit 107: Influence identification unit

Claims

1. A risk detection device for evaluating the impact of events on a business, comprising: an extraction unit that extracts keywords related to the event from a plurality of article data issued at different times; a generation unit that generates score time-series data indicating a temporal change in a risk score indicating the importance of the event based on the article data and the keywords; a calculation unit that calculates delay correlation data indicating a delay correlation, which is a correlation relationship considering the temporal deviation between the performance value and the risk score, based on business time-series data indicating a temporal change in a performance value related to the performance of the business and the score time-series data; an output unit that outputs notification data based on the delay correlation data.

2. The risk detection device according to claim 1, wherein the generation unit generates the score time-series data based on the number of occurrences of the keywords that appear in the article data for each time.

3. The generation unit generates the score time-series data based on the number of occurrences of the keywords related to each perspective on the event for each perspective, and the calculation unit calculates the delay correlation data for each perspective.

4. There are a plurality of types of the perspectives, and the generation unit generates the score time-series data for each perspective for each of the plurality of types.

5. The risk detection device according to claim 4, wherein the types of the perspectives include at least one of region, country, industry, and material.

6. For each article data, further comprising an addition unit that adds an industry-related tag to the article data based on a geographical location related to the content of the article data and industry geographical data indicating geographical characteristics that are geographical characteristics related to the industry, and the generation unit generates the score time-series data with the industry as the perspective based on the tag.

7. For each article data, further comprising an addition unit that adds a tag related to a category based on a category of an event related to a keyword extracted from the article data, and the generation unit generates the score time-series data with the category as the perspective based on the tag.

8. The delay correlation data includes a correlation coefficient representing the strength of the correlation between the actual value and the risk score, and a delay coefficient representing the magnitude of the time lag at which the correlation occurs between the actual value and the risk score. The risk detection device according to claim 1.

9. Based on the business time series data and the score time series data, the calculation unit calculates, for each perspective, a correlation coefficient representing the strength of the correlation between the actual value and the risk score, and a delay coefficient representing the magnitude of the time lag at which the correlation occurs between the actual value and the risk score, and calculates the correlation coefficient and the delay coefficient of the perspective where the correlation coefficient is equal to or greater than a predetermined value as the delay correlation data. The risk detection device according to claim 3.

10. The notification data includes information indicating a risk perspective, which is a perspective in which the risk score equal to or higher than the risk threshold exists, among the perspectives corresponding to the delay correlation data. The risk detection device according to claim 9.

11. The notification data includes a tag cloud indicating the keyword related to the risk perspective. The risk detection device according to claim 10.

12. The notification data includes a heat map indicating the degree of the risk score for each predetermined time unit. The risk detection device according to claim 1.

13. A risk detection method by a risk detection device for evaluating the impact of an event on a business, extracting a keyword related to the event from a plurality of article data issued at different dates and times, generating score time series data indicating the time change of a risk score indicating the importance of the event based on the article data and the keyword, calculating delay correlation data indicating a delay correlation, which is a correlation considering the time lag between the actual value and the risk score, based on business time series data indicating the time change of the actual value related to the performance of the business and the score time series data, outputting notification data based on the delay correlation data. A risk detection method.

14. On a computer for evaluating the impact of an event on a business a procedure for extracting a keyword related to the event from a plurality of article data issued at different dates and times, and a procedure for generating score time series data indicating the time change of a risk score indicating the importance of the event based on the article data and the keyword, A procedure for calculating delay correlation data indicating a delay correlation, which is a correlation considering the temporal deviation between the performance value and the risk score, based on the business time-series data showing the temporal change of the performance value related to the performance of the business and the score time-series data. A program for executing a procedure for outputting notification data based on the delay correlation data.

Images