Information processing system and information processing method

The information processing system accurately predicts task completion times by extracting time expressions from documents and using machine learning to analyze dynamic work schedules, addressing inaccuracies in existing methods.

JP2024163498APending Publication Date: 2024-11-22HITACHI LTD
View PDF 2 Cites 0 Cited by

Patent Information

Application Number
JP2023079182
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-12
Publication Date
2024-11-22

AI Technical Summary

Technical Problem

Existing methods for predicting task completion times are inaccurate due to reliance on worker-inputted values and inconsistent registration of progress and schedule information, lacking the ability to utilize detailed and dynamic information from document data.

Method used

An information processing system that extracts time expressions from business documents, estimates process engagement times, and predicts task completion times using machine learning models and natural language processing to analyze document data.

Benefits of technology

Enables highly accurate and automated prediction of task completion times based on the execution status described in document data, improving efficiency and reducing reliance on individual worker inputs.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024163498000001_ABST
    Figure 2024163498000001_ABST
Patent Text Reader

Abstract

To automatically and highly accurately predict a completion period of work on the basis of an implementation state etc., of the work written in document data, in an information processing system.SOLUTION: An information processing system comprises: a text extraction unit which extracts a sentence followed by time expression from a work document; a process attendance period estimation unit which identifies a process of work from the sentence on the basis of process related information associated with the work process, and refers to the time expression to estimate a process attendance period for each process of the work; and a work completion period analogy unit which predicts a completion period of the work on the basis of a work completion period parameter and the process attendance period.SELECTED DRAWING: Figure 1A
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

[0001] The present invention relates to an information processing system and an information processing method. [Background technology]

[0002] Predicting the completion date of a task with high accuracy leads to more efficient management and subsequent tasks, and prevents time and financial losses. For example, in system development work, if the completion date of development can be predicted from the daily work content, appropriate management based on the predicted value can be carried out, preventing delivery delays.

[0003] In sales activities aimed at receiving orders for product or system development, predicting the completion date of the order contract not only enables progress management of sales activities, but also makes it possible to formulate plans in advance for the allocation of development personnel and ordering of parts required after the contract is acquired, thereby improving management efficiency and shortening manufacturing and development times.

[0004] Existing technologies for predicting when a task is completed use labels or numerical information such as values ​​that indicate the characteristics of the target task, progress, and schedules, and do not utilize document data such as reports and minutes that are generated during the course of the task.

[0005] Predicting the completion time of a task based on values ​​that indicate the characteristics of the target task is difficult because it is not possible to make a prediction based on the actual status of the task. In addition, predictions using progress and schedules are left to the discretion of the worker, so the accuracy of the registered information and the frequency of registration are highly dependent on the individual, making it difficult to make a highly accurate prediction.

[0006] On the other hand, the contents of document data such as reports and minutes are thought to contain important information that can be used as a clue to predicting the completion time of tasks, such as detailed task execution status and the execution schedule that changes from time to time accordingly. There is a need for an automatic and highly accurate prediction of task completion time that takes into account the inherent information while mechanically processing this document data.

[0007] Related techniques include, for example, Patent Document 1, Patent Document 2, and Non-Patent Document 1.

[0008] Patent Document 1 discloses a method for predicting the required completion date for each process assuming the manufacturing and delivery work of infrastructure products such as elevators and storage.

[0009] Patent document 2 discloses a technology that selects reports that include specific behavioral indicators, extracts phrases indicating actions taken against customers as well as the dates and stores on which the actions were taken, and uses the dates and stores to determine the success or failure of the behavioral indicators.

[0010] Non-Patent Document 1 discloses a method for predicting the number of days of delay in a schedule set in an EPC (engineering, procurement, and construction) contract in the shipbuilding industry, from clues based on the frequency of specific words written in the contract documents at the time of signing. [Prior art documents] [Patent documents]

[0011] [Patent Document 1] JP 2019-145007 A [Patent Document 2] JP 2017-191442 A [Non-patent literature]

[0012] [Non-Patent Document 1] Byung-Yun Son and Eul-Bum Lee.2019. “Using Text Mining to Estimate Schedule Delay Risk of 13 Offshore Oil and Gas EPC Case Studies During the Bidding Process”Energies 12,no.10:1956 Summary of the Invention [Problem to be solved by the invention]

[0013] The clues used in Patent Document 1 for prediction are values ​​indicating the characteristics of the project and completed processes entered by the worker, and it does not disclose a method for predicting the completion time of a task based on the detailed status of work execution using document data or the associated changeable schedule of work execution.

[0014] Patent Document 2 extracts the activity content and the date of the activity content described in the document data, but does not disclose a method for using the activity content and the date of the activity content to predict the time of completion of the task.

[0015] Non-Patent Document 1 makes predictions only based on the frequency of specific words in documents created at the beginning of the contract, and does not disclose a method for predicting the number of days of delay based on the detailed work implementation status described in the document data or the work implementation schedule that changes from time to time accordingly.

[0016] An object of the present invention is to automatically and highly accurately predict the completion time of a task based on the execution status of the task described in document data in an information processing system. [Means for solving the problem]

[0017] An information processing system of one embodiment of the present invention is an information processing system that predicts a completion time of a task from a business document having sentences including time expressions related to the task, and is characterized by having: a text extraction unit that extracts sentences including the time expressions from the business document; a process engagement time estimation unit that identifies a process of the task from the extracted sentences based on process-related information related to the process of the task and estimates a process engagement time for each process of the task by referring to the time expressions; and a task completion time analogy unit that predicts the completion time of the task based on a task completion time parameter related to the completion time of the task and the process engagement time. Effect of the Invention

[0018] According to one aspect of the present invention, in an information processing system, it is possible to automatically and highly accurately predict the completion time of a task based on the progress status of the task described in document data. [Brief description of the drawings]

[0019] [Figure 1A] FIG. 1 is a diagram illustrating a configuration of an information processing system according to a first embodiment. [Figure 1B] FIG. 11 is a diagram illustrating a configuration of an information processing system according to a second embodiment. [Figure 1C] FIG. 11 is a diagram illustrating a configuration of an information processing system according to a second embodiment. [Figure 1D] FIG. 11 is a diagram illustrating a configuration of an information processing system according to a third embodiment. [Diagram 2] 13 is a flowchart showing a process of a text extraction unit. [Diagram 3] FIG. 11 is a diagram showing an example of a business document data list. [Figure 4] FIG. 11 is a diagram showing an example of an extracted data list. [Figure 5A] 13 is a flowchart showing the process of a process engagement timing estimation unit. [Figure 5B] 13 is a flowchart showing the process of a process engagement timing estimation unit. [Figure 6A] FIG. 13 is a diagram showing an example of an extracted data list with processes. [Figure 6B] FIG. 13 is a diagram showing an example of an extracted data list with processes. [Figure 7A] FIG. 13 is a diagram showing an example of an engagement time estimation result. [Figure 7B] FIG. 13 is a diagram showing an example of an engagement time estimation result. [Figure 8A] 13 is a flowchart showing a process of a task completion time analogy unit. [Figure 8B] 13 is a flowchart showing a process of a task completion time analogy unit. [Figure 9A] FIG. 11 is a diagram illustrating an example of reference data. [Figure 9B] FIG. 11 is a diagram illustrating an example of reference data. [Figure 10]FIG. 13 is a diagram illustrating an example of a prediction result. [Figure 11] 13 is a flowchart showing a process of a progress status estimation unit. [Figure 12] FIG. 11 is a diagram showing an example of a progress status estimation result. [Figure 13] FIG. 4 is a diagram showing an example of the content displayed on a display unit. [Figure 14] FIG. 2 is a diagram showing an example of the contents of a process-related information database. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0020] An embodiment will now be described with reference to the drawings. In this embodiment, one or more business documents are input for a specific business, and the completion time of the business is estimated and output. EXAMPLES

[0021] The configuration of an information processing system according to the first embodiment will be described with reference to Fig. 1A. In the first embodiment, the case of estimating the completion time of a contract for an order contract will be described as an example, with a sales activity aimed at winning a contract for an individual made-to-order product or system development from a customer as the target business.

[0022] The information processing system of the first embodiment predicts the completion time of a task from a task document having a sentence including a task process and a time expression related to the task.

[0023] As shown in FIG. 1A, the information processing system of the first embodiment includes a text extraction unit (101) that extracts sentences including time expressions from a business document, a process engagement time estimation unit (102) that identifies a business process from the extracted sentences based on process-related information related to the business process and estimates a process engagement time for each business process by referring to the time expressions, and a task completion time analogy unit (103) that predicts the completion time of the task based on a task completion time parameter (105) related to the completion time of the task and the process engagement time.

[0024] Furthermore, the information processing system of the first embodiment has a process related information database (104) that stores a list of multiple process steps as the process related information. The process engagement time estimation unit (102) identifies the process step from the text by comparing the extracted text with the list of process steps.

[0025] An example of the contents of the process related information database (104) is shown in FIG.

[0026] 14, a list of multiple work processes is stored in the process-related information database (104). The list of work processes includes, for example, "contact," "proposal activity," "quotation activity," and "preliminary notice and contract."

[0027] The task completion time parameter (105) is data including a set value that is the basis of the task completion time analogy processing, or the task completion time parameter (105) is data including a set value that controls the task completion time analogy processing.

[0028] For example, in a Support Vector Machine or a neural network, the types of formulas include coefficients that form the basis of a numerical calculation formula, and kernel functions and activation functions used therein.

[0029] In addition, in the case of RandomForest or the like, focusing on the output value of the process engagement time estimation unit (102), the control setting values ​​include the procedure for classifying the input data, the condition branch for classification, and the threshold value used in the condition.

[0030] In addition, it is a setting value in processing such as when a question created using the output of the process engagement time estimation unit (102) is given to a trained natural language generation model to obtain an answer. For example, in processing such as when the natural language generation model answers the question "If it takes X days to complete the proposed activity and Y days to complete the estimate, how many days do you think it will take to get an order contract?" with "Z days," the task completion time parameter (105) is a numerical value of an internal calculation formula in the natural language generation model and a template of a question.

[0031] The text extraction unit (101) detects and outputs sentences containing time expressions from business documents. The processing of the text extraction unit (101) will be described with reference to FIGS.

[0032] In S201, the text extraction unit (101) accepts a business document data list (301) consisting of pairs of business documents (304) and the writing dates (303) of the business documents. However, the business documents (304) may be any text data consisting of one or more words. Also, the writing dates (303) do not have to be the writing dates as long as they correspond to the business documents (304).

[0033] For example, in the case of a sales activity task targeted in the first embodiment, a business document data list (301) can be accepted that is composed of a sales activity report as a business document (304) and the date on which the sales activity was performed as the entry date (303).

[0034] In S202, processing is started for each business document data (302) held in the input business document data list (301). In S203, the business document (304) is divided into sentence data (403). Here, the sentence data (403) is text data consisting of one or more words. In the division into sentence data (403) (S203), the sentences are divided by commas.

[0035] However, the splitting method in S203 may be a rule- or pattern-based method such as splitting at periods or line breaks or splitting based on regular expressions, a method using a statistical model, a machine learning or deep learning model, or a combination of both.

[0036] Furthermore, the division unit of the sentence data (403) does not have to be a sentence, and any unit may be adopted, such as a paragraph unit, a clause unit, a specific number of words, etc. In S205, it is determined whether a time expression exists in each sentence data (403), and if a time expression exists, it is stored as a time expression (404) together with the sentence data (403) in the extracted data (402).

[0037] However, the time expression is an expression that specifies the occurrence time or occurrence interval of an arbitrary event on a time axis. In the first embodiment, the time expression (404) is an expression that indicates a position on the time axis at a date granularity. For example, January 25, 6, and around the 15th, early July, and the end of August, 2020. If the sentence data (403) contains a time expression (404), in S206, the business document data (302) to be processed is stored in the extracted data (402), and the extracted data (402) is added to the extracted data list (401).

[0038] However, if the sentence data (403) contains multiple time expressions (404), the sentence data (403) and the business document data (302) are assigned to each time expression (404) and added to the extracted data list (401). In the first embodiment, the process (S205) of determining the presence of a time expression (404) is performed by pattern matching using a regular expression.

[0039] For example, a regular expression such as "((1[0-9])|(2[0-9])|([30|1])|[1-9]) days" can detect around January 25, 6, and 15, 2020. However, the presence determination process may be performed on a rule basis, or may be performed using a statistical model, a machine learning or deep learning model, or a combination of both.

[0040] After the above process is applied to all business document data (302), an extracted data list (401) is output (S209), and the process of the text extraction unit (101) is completed.

[0041] The process engagement time estimation unit (102) receives the extracted data list (401) output from the text extraction unit (101) and outputs an estimation result of the engagement time of each work process. The process engagement time estimation unit (102) will be described with reference to Figs. 5A, 6A, and 7A.

[0042] In S501, the process engagement timing estimation unit (102) receives the extracted data list (401).

[0043] In S502, the process engagement time estimation unit (102) starts processing each extracted data (402) in the extracted data list (401). In S503, the time expression (404) held by the extracted data (402) is normalized to an arbitrary format expressing the corresponding date. However, normalization may be performed using a longer or shorter time unit such as a year, month, and time, not limited to the date.

[0044] In the first embodiment, the time information (404) is normalized to the format of "year-month-day" (e.g., 2022-06-12). However, if the date cannot be identified from the time expression (404), the unidentified date is filled with an arbitrary tag indicating that it is a missing value. For example, a time expression of "mid-June" is converted to a format such as "<missing>-06-<missing>". In S504, the missing value of the time expression (404) normalized in S503 is complemented with a specific value based on the sentence data (403) and business document data (302) held in the extracted data (402).

[0045] For example, in "<missing>-06-<missing>", the missing year is assigned the same year as the entry date (303) of the original business document, and the missing day is assigned the value "15" based on the expression "mid-month" as the complementary value. In addition, in the case of the time expression (404) "end of January", if the entry date (303) of the original business document data (302) is between January and July, the missing year is assigned to the year of the entry date (303), and if the entry date (303) of the original business document data (302) is between August and December, the missing year is assigned to the year of the entry date (303) plus 1 as the complementary value.

[0046] In addition, based on the expression "end" in the time expression (404), the missing day is assigned a value of "31". In the case of the time expression (404) of "on the 25th", the missing year and month are assigned the year and month of the entry date (303) of the original business document data (302) as a complement value. However, the complementation process (S504) may be performed on a rule basis, or may be performed using a statistical model, machine learning or deep learning model, or may be performed by combining these means. Furthermore, the complementation process (S504) may be performed by referring to external data and using information not held in the extracted data (402).

[0047] In S505, the normalized time expression (605) is compared with the writing date (303) of the business document data (302) to determine whether the matter described in the sentence data (403) in the extracted data (402) is an implementation item or a planned item. If the date indicated by the normalized time expression (605) is later than the writing date (303), the category (606) of the extracted data (402) is set to the planned date. If the date indicated by the normalized time expression (605) is the same as the writing date (303) or earlier than the writing date (303), the category (606) of the extracted data (402) is set to the implementation date. In S506, the related process (607) of the extracted data (402) is determined.

[0048] In determining the related processes (S506), the process related information database (104) in Fig. 14 is referenced. As described above, the process related information database (104) is a database that accumulates information related to the processes of a business. In addition to the processes of a business, the process related information database (104) can accumulate various types of information, such as a dictionary indicating a determination method for linking a predefined process with the extracted data (402), learned parameters of a machine learning model that classifies the sentence data (403) into one of the predefined processes, and a collection of documents related to a business for which a process classification is not defined in advance.

[0049] In the first embodiment, as shown in Fig. 14, an example of determining related steps (607) using a dictionary is shown using four steps defined in advance as steps of sales activity business: "Contact", "Proposal activity", "Estimate activity", and "Unofficial notice / contract". However, the number of steps may be any number greater than or equal to 1.

[0050] If a specific character string exists in the sentence data (403) of the extracted data (402), the process corresponding to the extracted data (402) is assigned as a related process (607).

[0051] The character strings for the “Contact” Step are “kickoff,” “hearing,” “inquiry,” etc. The character strings for the “Proposal Activities” Step are “proposal,” “RFP,” etc. The character strings for the “Estimation Activities” Step are “Estimation,” “Requirements Definition,” etc. The character strings for the “Unofficial Notice / Contract” Step are “Unofficial Notice,” “Contract,” “Received Order,” “Order,” “Purchase Order,” “Inspection,” etc. If none of the character strings exist in the sentence data (403), the related step (607) given to the extracted data (402) is “none.”

[0052] In S507, the normalized time expression (605) calculated in S503 to S506, the classification (604) of the extracted data (402), the related process (607), the original extracted data (402), and the business document data (302) from which the extracted data (402) was extracted are combined to generate step-attached extracted data (602), and the step-attached extracted data (602) is added to the step-attached extracted data list (601). The above process is applied to all extracted data (402) in the extracted data list (401), and the loop process of S502 ends.

[0053] In S508, the process engagement time (704) for each process is determined using the process-attached extracted data list (601). In S508, first, the process-attached extracted data (602) corresponding to each process is aggregated. Among the process-attached extracted data (602) aggregated for each process, data with low reliability as the date of engagement for the process is excluded. The above-mentioned exclusion process may be based on a specific rule, or may be performed using a statistical model, a machine learning / deep learning model, or the like, or may be performed by combining them.

[0054] In the first embodiment, step S508 targets the process-attached extracted data (602) whose category (604) is a scheduled date and excludes, for each process type, all process-attached extracted data (602) other than the business document data whose entry date (303) is the furthest in the future.

[0055] In the first embodiment, the process-attached extracted data (602) whose classification (604) is an implementation date is not excluded. The process-attached extracted data (602) remaining after the above-mentioned exclusion process is used to calculate the process engagement time (704) for each type of process. The process engagement time (704) may be calculated based on rules, or may be calculated using a statistical model, a machine learning / deep learning model, or the like, or may be calculated by combining them.

[0056] In addition, if the form of the calculated process engagement time (704) is information regarding time, it may take various forms, such as a single value such as a specific date or month, a range of dates or months, a probability distribution of dates or months, or a character string indicating the process engagement time (704).

[0057] In the first embodiment, the end date (705) of a process is determined on a rule-based basis for each process. The related processes (607) and normalized time expressions (605) of the process-attached extracted data (602) remaining after the exclusion process are referenced, and the most future normalized time expression (605) is obtained for each process, and this is set as the end date (705) of the process. The end date (705) of the process is combined with the process, and stored as the process engagement time (704).

[0058] In S509, the ID (703) of the task to be processed, the process engagement time (704) calculated in S508, and the processing completion date and time of S508 are aggregated as the estimated processing implementation time (702) into an engagement time estimation result (701), and the engagement time estimation result (701) is output. After the above processing, the processing of the process engagement time estimation unit (102) ends.

[0059] The task completion time analogy unit (103) receives the task engagement time estimation result (701) output from the task engagement time estimation unit (102) and the task completion time parameter (105), and outputs the task completion time analogy result. The process of the task completion time analogy unit (103) will be described with reference to Figs. 8A, 9A, and 10.

[0060] In S801, the process engagement time estimation result (701) is accepted. In S802, the difference between 23:59:59 on the end date (705) of each process in the process engagement time estimation result (701) and the estimated processing implementation time (702) is calculated, and the number of days required for the process (905) is calculated. The calculated number of days required for each process (905) is used as feature amount data (904) and is combined with the task ID (703) and estimated processing implementation time (702) of the process engagement time estimation result (701) to create reference data (901). However, any numerical value may be added to the feature amount data (904).

[0061] For example, the number of days difference between the end dates of processes, which can be calculated from the end dates (705) of each process, or task information may be obtained as a numerical value by referencing an external database based on the task ID (902) of the reference data (901), and included in the feature data (904).

[0062] In the first embodiment, not only the number of days required by each process (905) but also the expected order date (expected completion date of winning the order contract) entered by the sales representative is obtained from the task information linked to the task ID (902), the difference in days between the expected order date at 11:59:59 and the estimated processing execution time (702) is calculated, and the difference in days is added to the feature data (904) as the number of days until the expected order is received (906). In S803, the task completion time is inferred. In the first embodiment, the feature data (904) is used as an input, and the number of days required until the task completion (1203) is predicted using a machine learning model such as SupportVectorMachine or RandomForest.

[0063] However, the analogy process may be a rule or pattern-based method, a method using a statistical model other than SupportVectorMachine or RandomForest, a machine learning or deep learning model, or a combination of both. In the first embodiment, the parameters of SupportVectorMachine and RandomForest are read from the task completion time parameter (105).

[0064] The task completion time parameters (105) are data that store various parameters used for inferring the task completion time (S803). However, the task completion time parameters (105) are not limited to machine learning parameters such as SupportVectorMachine and RandomForest, but may also include various parameters required for rule-based inference processing and deep learning parameters.

[0065] Furthermore, the parameters do not necessarily have to exist in the same server as the information processing system that performs the analogy processing in S803. The parameters of SupportVectorMachine and RandomForest used in the first embodiment are learned in advance using feature data (904) of past completed tasks and the number of days required to complete the task (1203). The task completion time (1204) is calculated by adding the predicted number of days required to complete the task (1203) to the estimated processing execution time (903) of the reference data (901). The task ID (902) of the reference data (901) is combined with the created number of days required (1203) and the task completion time (1204) to create a prediction result (1201). In S804, the calculated prediction result (1201) is output. After the above processing, the processing of the task completion time analogy unit (103) ends. According to the first embodiment, it is possible to automatically and highly accurately predict the time when a task will be completed by using the time expressions of the related steps required to complete the task described in the business document as a clue. EXAMPLES

[0066] In the second embodiment, the embodiment of Figures 1B and 1C will be described using an example of estimating the completion time of supplier selection in a procurement activity aimed at selecting a supplier of parts required for product manufacturing in the manufacturing industry.

[0067] 1B, a progress estimation unit (106) estimates the progress of a task based on the contents of the text extracted by the text extraction unit (101). A task completion time analogy unit (103) predicts the completion time of the task by referring to the progress.

[0068] In FIG. 1C, a progress estimation unit (106) estimates the progress of work based on the content of the text extracted by the text extraction unit (101) and the process-related information stored in the process-related information database (104).

[0069] The text extraction unit (101) detects sentences containing time expressions from business documents and outputs them. The processing of the text extraction unit (101) will be explained using Figures 2, 3, and 4. The text extraction unit (101) receives a business document data list (301) consisting of pairs of business documents (304) and the dates (303) on which the business documents were written.

[0070] In the second embodiment, the minutes of meetings of the staff members working on the target procurement task, as well as the chat history and meeting minutes of those staff members with suppliers, are treated as business documents (304), and the dates on which the chat history and meeting minutes were created are treated as the writing dates (303).

[0071] In S202, processing is started for each business document data (302) held in the input business document data list (301). In S203, the business document (304) is divided into sentence data (403). In the division into sentence data (403) (S203), the sentences are divided using spaces, line breaks, etc. In S205, it is determined whether a time expression is present in each sentence data (403), and if a time expression is present, it is stored as a time expression (404) together with the sentence data (403) in the extracted data (402).

[0072] If the sentence data (403) contains a time expression (404), the business document data (302) to be processed is stored in the extracted data (402) in S206, and the extracted data (402) is added to the extracted data list (401). However, if the sentence data (403) contains multiple time expressions (404), the sentence data (403) and the business document data (302) are assigned to each time expression (404) and added to the extracted data list (401).

[0073] In the second embodiment, the process (S205) for determining the presence of the time expression (404) is performed by pattern matching using a regular expression. For example, the dates January 25, 6, and 15, 2020 can be detected using a regular expression such as "((1[0-9])|(2[0-9])|([30|1])|[1-9]) days." After applying the above process to all business document data (302), an extracted data list (401) is output (S209), and the process of the text extraction unit (101) ends.

[0074] The process engagement time estimation unit (102) receives the extracted data list (401) output from the text extraction unit (101) and outputs an estimation result of the engagement time of each work process. The process engagement time estimation unit (102) will be described with reference to Figs. 5B, 6B, and 7B.

[0075] In S522, processing is started for each extracted data (402) in the extracted data list (401). In S523, the time expression (404) held by the extracted data (402) is normalized to an arbitrary format expressing the corresponding date and time.

[0076] In the second embodiment, the time information (404) is normalized to a format of "year-month-day-time" (e.g., 2022-06-12-17). However, if the date or time cannot be identified from the time expression (404), the unidentified date or time is filled with an arbitrary tag indicating that it is a missing value. For example, a time expression such as "around 5 p.m. on the 12th" is converted to a format such as "<missing>-<missing>-12-17".

[0077] In S524, the missing values ​​of the time expression (404) normalized in S523 are complemented with specific values ​​based on the sentence data (403) and business document data (302) held in the extracted data (402). The complementation process for the year, month, and day is the same as in the first embodiment. For missing time, "23" is complemented.

[0078] However, the complementation process (S524) may be performed based on rules, or may be performed using a statistical model, a machine learning or deep learning model, or may be performed by combining them. Also, the complementation process (S524) may be performed by referring to external data and using information not held in the extracted data (402).

[0079] In S526, a related process (627) of the extracted data (402) is determined. In determining the related process (S526), ​​the process related information database (104) is referenced. In the second embodiment, the process related information database (104) has parameters of a deep learning model that identifies a process related to the sentence data (403).

[0080] The parameters may be parameters learned in advance from data, or may be parameters set manually. For example, supervised learning may be performed in advance from data to which the relevant related steps (627) are assigned to the sentence data (403), or a clustering model may be learned in advance to divide a plurality of sentence data (403) to which the related steps (627) are not assigned into groups equal to the number of defined related steps.

[0081] In S526, the parameters of the deep learning model are read, and the related steps (627) for each piece of sentence data (403) are predicted by the deep learning model. In the second embodiment, the related steps (627) output by the deep learning model are "investigation," "contact," "negotiation," "contract," and "not related." However, the number of types of related steps (627) may be any number as long as they are one or more.

[0082] In S527, the normalized time expression (625) calculated in S523 to S526, the related process (627), the original extracted data (402), and the business document data (302) from which the extracted data (402) was extracted are combined to create extracted data with processes (622), and the extracted data with processes (622) is added to the extracted data with processes list (621). The above processing is applied to all extracted data (402) in the extracted data list (401), and the loop processing in S522 ends.

[0083] In S528, the process engagement time (724) of each process is determined using the process-attached extracted data list (621). In S528, first, the process-attached extracted data (622) corresponding to each process is aggregated. From the process-attached extracted data (622) aggregated for each process, the end date (725) of the process is obtained as the process engagement time (724). In the second embodiment, the probability that the process will end is calculated for each month, and this is set as the end month probability (726). However, the calculation of the probability may be performed on a rule basis, or may be performed using a statistical model, a machine learning / deep learning model, or the like, or may be performed by combining them.

[0084] In the second embodiment, a deep learning model is used that receives as input all sentence data (623) of the process-specific extracted data (622) aggregated for each process, and outputs 25 probability values ​​indicating which month, from 12 months before to 12 months ahead, is the month in which the process will end. The parameters of this deep learning model are read from the process-related information database (104). The probability is set as the end month probability (726), and the last day of the month with the highest probability among them is set as the end date (725) of the process.

[0085] The end date (725) of the process and the probability (726) of the end date are stored as the process engagement time (724). In S529, the ID (723) of the task to be processed, the process engagement time (724) calculated in S528, and the processing completion date and time of S528 are aggregated as the estimated processing implementation time (722) into the engagement time estimation result (721), and the engagement time estimation result (721) is output. After the above processing, the processing of the process engagement time estimation unit (102) ends.

[0086] The progress estimation unit (106) receives the extracted data list (401) output from the text extraction unit (101) and outputs an estimation result of the progress of the work. The process of the progress estimation unit (106) will be described with reference to Figs. 3, 4, 5B, 6A, 11, and 12.

[0087] In the second embodiment, the smoothness of the business is calculated in S1310, but this is optional, so the reception process of the business document data in S1309 is not a mandatory requirement.

[0088] In the processes from S1301 to S1307, a process-attached extracted data list (601) is obtained, but this process is similar to the description of S501 to S507 of the process engagement timing estimation unit (102) in the first embodiment, so it will be omitted. However, unlike S506 in the first embodiment, in S1306, the related process (607) for each sentence data (403) is predicted by a deep learning model, and the related process (607) is "investigation," "contact," "negotiation," "contract," and "not related." However, the type of related process (607) may be any number of types as long as it is one or more. The parameters of the deep learning model are read from the process related information database (104).

[0089] In S1308, the current process is determined. Process-attached extracted data (602) having a normalized time expression (605) that is the same as or earlier than the execution date of S1308 is aggregated, and the process that is closest to the completion of the business among the related processes (607) of the process-attached extracted data (602) is set as the current process (1405). In S1309, the business document data list (301) is accepted. In S1310, the smoothness of the business is calculated.

[0090] In the second embodiment, the ratio of positive sentences is calculated for all business documents (304) in the business document data list (301), and this is taken as the smooth progress of business (1406). The ratio of positive sentences is calculated by classifying sentences into two values ​​according to whether they have positive or negative content. The classification process may be performed on a rule basis, or may be performed using a statistical model, a machine learning or deep learning model, or a combination of these.

[0091] In S1311, the calculated current step (1405), the smoothness of the task (1406), the ID (1403) of the task to be processed, and the processing date and time of S1311 are used as the estimated processing execution time (1402) to combine the above information to calculate a progress status estimation result (1401). The progress status estimation result (1401) is output, and the processing of the progress status estimation unit (106) ends.

[0092] The task completion time analogy unit (103) receives the process engagement time estimation result (721) output from the process engagement time estimation unit (102), the progress status estimation result (1401) output from the progress status estimation unit (106), and the task completion time parameter (105), and outputs the task completion time analogy result. The processing of the task completion time analogy unit (103) will be described with reference to Figs. 7B, 8B, 9B, 10, and 12.

[0093] In S821, the process engagement time estimation result (721) is received. In S822, the progress status estimation result (1401) is received. In S823, the task start date (924) is obtained from the task information corresponding to the task ID (723), and the number of days required by each process (926) is calculated by taking the difference between the end date (725) of each process in the process engagement time estimation result (721) (23:59:59) and the task start date (924).

[0094] However, to calculate the number of days required by process (926), any time of the estimated processing execution time (722) may be used instead of the task start date (924), or any date obtainable from task information corresponding to the task ID (723) may be used. The calculated number of days required by process (926), the current process (1405) of the progress status estimation result (1401), and the smoothness of the task (1406) are combined to create feature data (925), and the task ID (723), estimated processing execution time (722), and task start date (924) of the task engagement timing estimation result (721) are added to create reference data (921).

[0095] However, any numerical value may be added to the feature amount data (925). In the second embodiment, the difference in days between the end dates (725) of each process is calculated for each combination of processes, and the difference in days is included in the feature amount data (925) as the number of days between processes (927).

[0096] In S824, the time of task completion is inferred. In the second embodiment, the feature data (904) is input, and the number of days required to complete the task (1203) is predicted using a deep learning model such as a Feedforward Neural Network with overlapping linear layers. However, the deep learning model may be any deep learning model such as a Convolutional Neural Network (CNN) or a Recurrent Neural Network (RNN), and the inference process in S824 may be a rule or pattern-based method, a method using a statistical model or machine learning such as Support Vector Machine or Random Forest, or a combination of these.

[0097] In the second embodiment, the parameters of the deep learning model are read from the task completion time parameters (105). The parameters of the deep learning model used in the second embodiment are learned in advance using feature data (925) of past completed tasks and the number of days required to complete the task (1203). The task completion time (1204) is calculated by adding the predicted number of days required to complete the task (1203) to the estimated processing execution time (923) of the reference data (921). The task ID (922) of the reference data (921) is combined with the generated number of days required (1203) and task completion time (1204) to create a prediction result (1201).

[0098] In S825, the calculated prediction result (1201) is output. After the above processing, the processing of the task completion time analogy unit (103) ends.

[0099] According to the second embodiment, it is possible to automatically and accurately predict the time when a task will be completed by using the time expressions of the related steps required to complete the task described in the business document as a clue. EXAMPLES

[0100] In the third embodiment, the embodiment of Fig. 1D having a display unit will be described by taking as an example a case where a sales activity aimed at obtaining a contract for an individual made-to-order product or system development from a customer is the target business and the completion time of the contract is estimated. However, in the third embodiment, only the display unit (107) will be described using Figs. 4, 6A, 7A, 10, 12, and 13.

[0101] The display unit (107) receives as input the extracted data list (401) output by the text extraction unit (101), the engagement time estimation result (701) output by the process engagement time estimation unit (102), the process-attached extracted data list (601) created by the process engagement time estimation unit (102), the progress status estimation result (1401) output by the progress status estimation unit (106), and the prediction result (1201) output by the task completion time analogy unit (103), and outputs the UI shown in Fig. 13. The prediction result (1501) of the task completion time by the system uses the value of the task completion time (1204) of the prediction result (1201).

[0102] The current process (1502) uses the value of the current process (1405) from the progress status estimation result (1401). The sales activity progress (1503) uses the value of the business progress (1406) from the progress status estimation result (1401). The process engagement time estimate (1504) is created based on the value of the process engagement time (704) from the engagement time estimation result (701).

[0103] In the third embodiment, a Gantt chart is created for each process, with the months between the start date and end date (705) of the process colored based on the end date (705) of the process. In the third embodiment, the start date of each process is set to the end date (705) of the previous process and colored. In addition, in the process-attached extracted data list (601) for each process whose category (606) is "implementation date", the month corresponding to the most future normalized time expression (605) is calculated for each type of related process (607), and the Gantt chart up to that month is colored as "implemented" for each type of related process (607).

[0104] However, the start date may be determined using a separate rule-based process, a statistical model, machine learning, or deep learning. Also, coloring based on the category (606) is not required. Also, the display format does not have to be a Gantt chart. The sentence data (1505) for each process is displayed by dividing the sentence data (603) of the process-attached extracted data (602) by type of related process (607).

[0105] The execution items (1506) display sentence data (603) of the step-attached extracted data for each step whose category (606) is "execution date." The plan items (1507) display sentence data (603) of the step-attached extracted data for each step whose category (606) is "scheduled date."

[0106] According to the third embodiment, it is possible to automatically and highly accurately predict the time when a task will be completed, based on the time expressions of related processes required to complete the task described in the business document. In addition, it is possible to manage the progress and actual / budget management of sales activities based on the mechanical analysis results shown on the display unit, which prevents reliance on individual personnel and improves business efficiency.

[0107] Here, the information processing system shown in FIG. 1A, FIG. 1B, FIG. 1C, and FIG. 1D is configured, for example, with an input device, an output device, a memory, a storage device, and a computer having a CPU.

[0108] Moreover, the functions of the "units" shown in FIGS. 1A, 1B, 1C, and 1D are realized by, for example, a processor (such as a CPU) executing a program.

[0109] For example, the text extraction unit (101) realizes a text extraction function by executing a program with a processor (CPU, etc.). The process engagement time estimation unit (102) realizes a process engagement time estimation function by executing a program with a processor (CPU, etc.). The task completion time analogy unit (103) realizes a task completion time analogy function by executing a program with a processor (CPU, etc.). The progress status estimation unit (106) realizes a progress status estimation function by executing a program with a processor (CPU, etc.).

[0110] According to the above embodiment, it is possible to automatically and accurately predict the time when a task will be completed by using clues such as the task execution status described in the document data and the schedule of the worker, which changes as the execution status changes.

[0111] In addition, since the time when tasks will be completed can be automatically predicted based on the results of mechanically analyzing the contents of the document data, this is expected to prevent dependency on individual tasks and improve work efficiency compared to a system in which the progress and schedule are separately entered by the worker. [Explanation of symbols]

[0112] 101 Text Extraction Unit 102 Process Engagement Time Estimation Department 103 Business completion time analogy section 104 Process-related information database 105 Task Completion Time Parameter 106 Progress Estimation Department 107 Display section

Claims

1. An information processing system for predicting a completion time of a task from a task document having a sentence including a time expression related to the task, comprising: a text extraction unit that extracts sentences including the time expression from the business document; a process engagement time estimation unit that identifies a process of the work from the extracted sentence based on process-related information related to the process of the work, and estimates a process engagement time for each process of the work by referring to the time expression; a task completion time analogy unit that predicts a completion time of the task based on a task completion time parameter related to a completion time of the task and the process engagement time; An information processing system comprising:

2. 2. The information processing system according to claim 1, The text extraction unit Dividing the business document into a plurality of said sentences; An information processing system comprising: a step of extracting sentences including said time expressions from said business documents by determining whether said time expressions exist in said divided sentences.

3. 2. The information processing system according to claim 1, The process engagement time estimation unit, normalizing the time expressions included in the extracted sentences; An information processing system comprising: a data processing unit for inputting missing dates and times into the normalized time representation;

4. 2. The information processing system according to claim 1, The process engagement time estimation unit, The time expressions included in the extracted sentences are divided into time expressions indicating work that has been performed and time expressions indicating future work plans; An information processing system that estimates the time to engage in the process using the classification.

5. 2. The information processing system according to claim 1, a process-related information database for storing a list of work processes of the plurality of work processes as the process-related information; The process engagement time estimation unit, An information processing system characterized in that the extracted sentence is compared with the list of work processes, thereby identifying the work process from the sentence.

6. 2. The information processing system according to claim 1, The task completion time analogy unit is An information processing system, comprising: an information processing unit for predicting a completion time of the task using data including a setting value for controlling an analogy process for inferring the completion time of the task as the task completion time parameter;

7. 2. The information processing system according to claim 1, The task completion time analogy unit is An information processing system characterized by predicting the completion time of the work by using the number of days required for each process of the work.

8. 2. The information processing system according to claim 1, The task completion time analogy unit is An information processing system that predicts the completion time of the task using a machine learning model.

9. 2. The information processing system according to claim 1, The task completion time analogy unit is An information processing system characterized by predicting the completion time of the task using a deep learning model.

10. 2. The information processing system according to claim 1, a progress status estimation unit that estimates a progress status of the work based on the content of the sentence extracted by the text extraction unit, The task completion time analogy unit is An information processing system that predicts the time when the work will be completed by referring to the progress status.

11. The information processing system according to claim 10, The progress status estimation unit, An information processing system that estimates the progress status by referring to the process-related information.

12. The information processing system according to claim 10, The progress status estimation unit, An information processing system characterized by estimating a process of the business currently being performed.

13. The information processing system according to claim 10, A display unit is provided. The display unit is An information processing system characterized by displaying at least one of the text extracted for each process of the work, a process engagement time estimate created based on the process engagement time, a prediction result for the completion time of the work, and an estimation result of the progress status of the work.

14. The information processing system according to claim 13, The display unit is displaying a currently performed process of the work as an estimated result of the progress of the work; An information processing system characterized in that the process engagement time estimate is divided into processes of the work that have been performed and processes of the work that are scheduled to be performed, and displayed for each process of the work.

15. 1. An information processing method for predicting a completion time of a task from a task document having a sentence including a time expression related to the task, comprising: a text extraction step of extracting sentences including the time expression from the business document; a process engagement time estimation step of identifying a process of the work from the extracted sentence based on process related information related to the process of the work, and estimating a process engagement time for each process of the work by referring to the time expression; A task completion time analogy step of predicting a completion time of the task based on a task completion time parameter related to the completion time of the task and the process engagement time; 13. An information processing method comprising:

Citation Information

Patent Citations

  • Information generation device and information generation method

    JP2017191442A

  • Project progress prediction device and project progress prediction system

    JP2019145007A