Data prediction system and data prediction method

Abstract

Accurately predict future survey response results from time series data of responses to surveys that have already been conducted. [Solution] A data prediction system disclosed herein uses time-series data of a desired item to calculate a predicted value of the desired item at a desired time (predicted value obtained by numerical calculation processing). The system also extracts a predetermined number of items as highly correlated items from multiple items, whose time-series data are highly correlated with the time-series data of the desired item, applies natural language processing to the predetermined number of highly correlated items, generates correction suggestion information for the predicted value, and corrects and outputs the predicted value based on the correction suggestion information.

Description

[Technical Field]

[0001] The present disclosure relates to a data prediction system and a data prediction method. [Background technology]

[0002] Traditionally, when a survey (questionnaire survey) is conducted in a certain district or region, the survey results (response results) are analyzed and used as marketing information or for product development. As a method for analyzing such questionnaire response results, for example, there is a technique described in Patent Document 1. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent Publication No. 2021-86443 Summary of the Invention [Problem to be solved by the invention]

[0004] However, conventional survey (questionnaire) result analysis techniques, such as those described in Patent Document 1, only provide analysis results of acquired survey responses and do not predict future survey response results. For example, when survey response results from 2018 to 2023 have been obtained for multiple survey items (multiple questionnaire items), it would be convenient for survey implementers if they could accurately predict the survey response results for 2024 without actually conducting the same survey. This is because being able to predict survey response results in this way would reduce the cost of conducting the survey and also enable the analysis of survey response results to be carried out quickly. In view of such circumstances, the present disclosure proposes a technique for accurately predicting data that is likely to be obtained in the future from acquired time-series data (for example, from the past to the present). [Means for solving the problem]

[0005] In order to solve the above problems, the present disclosure provides, as an example, A data prediction system that predicts future data based on time series data acquired over a predetermined period, a storage device that stores a program for performing time series prediction on a desired item among a plurality of items based on an aggregate value of time series data on the plurality of items; a processor that reads the program from the storage device and executes it; The processor: A process of calculating a predicted value of the desired item at a desired time using time series data of the desired item; a process of extracting a predetermined number of items having time series data highly correlated with the time series data of the desired item from the plurality of items as highly correlated items; the predetermined number of highly correlated items On a large scale language Input to the model and providing suggestion information for correction of the predicted value. Generate and a process of correcting the predicted value based on the correction suggestion information and outputting a corrected predicted value; We propose a data prediction system that performs the following.

[0006] Further features related to the present disclosure will become apparent from the description and accompanying drawings of this specification, and aspects of the present disclosure may be realized and realized by the elements and combinations of various elements and aspects set forth in the following detailed description and the appended claims. The descriptions herein are exemplary and illustrative only and are not intended to limit the scope or application of the present disclosure in any way. [Effects of the Invention]

[0007] According to the technology of the present disclosure, it is possible to accurately predict data that is likely to be obtained in the future from acquired time-series data (for example, from the past to the present). [Brief explanation of the drawings]

[0008] [Figure 1]FIG. 2 is a diagram for explaining the concept of data complementary prediction according to the present embodiment. [Figure 2] 1 is a diagram illustrating an example of a schematic configuration of a data interpolation prediction device 10 according to the present embodiment. [Figure 3] FIG. 1 is a diagram for explaining an outline of a data complementary prediction process according to the present embodiment. [Figure 4] 10 is a flowchart illustrating details of a data complement prediction process according to the present embodiment. [Figure 5] 10 is a table showing an example of a predicted value correction result by the data interpolation prediction process of the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0009] This embodiment proposes a data complementation prediction technology that predicts future data based on time series data acquired over a predetermined period. One of the features of the technology disclosed in this embodiment is that a predicted value calculated by numerical calculation using aggregate values ​​of desired items (items specified by a user) included in time series data accumulated in the past is corrected based on information on correction suggestions obtained by performing natural language processing on a group of other items that are highly correlated with the specified items, thereby providing a predicted value (after correction) that takes into account the current state of the world. Note that any type of data may be used as the time series data.

[0010] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. In the accompanying drawings, functionally identical elements may be designated by the same numerals. Note that the accompanying drawings illustrate specific embodiments and implementation examples according to the principles of the present disclosure, but these are for understanding the present disclosure and are not to be used to interpret the present disclosure in any way as being limiting.

[0011] Although the present embodiment has been described in sufficient detail to enable those skilled in the art to implement the present disclosure, it should be understood that other implementations and forms are possible, and that changes in configuration and structure and substitutions of various elements are possible without departing from the scope and spirit of the technical ideas of the present disclosure. Therefore, the following description should not be interpreted as being limited thereto.

[0012] <Concept of data complementary prediction> FIG. 1 is a diagram illustrating the concept of data complement prediction according to this embodiment. Data complement prediction involves predicting (calculating predicted values) future survey response results (specific items specified by the user) for a specific region using past survey response results (time-series data: for example, Type 1 survey data for the Kanto region; Type 1 survey data refers to survey data on personal preferences in a specific region) for the same region. The Type 1 survey targets individuals (men and women aged 12 to 69) residing in, for example, the Kanto region, Kansai region, Nagoya region, Northern Kyushu region, Sendai region, Hiroshima region, and Sapporo region (seven regions), and primarily surveys basic attributes such as gender, age, and annual income, media exposure (e.g., television, the Internet, and magazines), involvement with products and services, lifestyle behavior, shopping behavior, and attitudes and values. Although not shown in FIG. 1 , future survey response results for the same region (prefecture) may also be predicted using time-series data (survey response results for a specific prefecture over a specified number of years) of Type 2 survey data (a lifestyle survey of individuals in a specific prefecture) in addition to the Type 1 survey data. Here, the second type of survey involves, for example, conducting surveys to grasp the reading status of major newspapers in all 47 prefectures of the country, as well as the lifestyle behavior and actual consumption status of consumers.

[0013] As shown in Figure 1, in the data supplementary prediction according to this embodiment, for example, based on Type 1 survey data (such as the Kanto region or Kansai region) obtained each year from 2014 to 2023, the survey response results for the same region in 2024, even though no survey has actually been conducted, are predicted (predicted values ​​for specific survey items are calculated).

[0014] <Configuration example of data augmentation prediction device> FIG. 2 is a diagram showing an example of a schematic configuration of a data interpolation prediction device 10 according to this embodiment.

[0015] (i) The data complement prediction device 10 can be configured using a general-purpose computer, and includes a processor (control device) 101 such as a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or a GPU (Graphics Processing Unit), a storage device 102, a communication device 103, an input device 104, and an output device 105.

[0016] (ii) The storage device 102 is composed of a ROM (Read Only Memory), a RAM (Random Access Memory), and an HDD (Hard Disk Drive), and stores a program for data complementation prediction, the results of responses to Type 1 surveys that have already been conducted (Type 1 survey data), the results of responses to Type 2 surveys that have already been conducted (Type 2 survey data), and Large-scale language Model (e.g., GPT-4o) etc.

[0017] The processor 101 reads a program for data complementary prediction from the storage device 102 and expands it in an internal memory (not shown), thereby constructing a prediction value calculation unit 1011, a correlation calculation unit 1012, and a prediction value correction unit 1013.

[0018] The communication device 103 is used, for example, to receive data (each of the past type 1 survey data and each of the past type 2 survey data) transmitted from a computer in a remote location, or to communicate with another data complement prediction device 10' (not shown) and obtain its data complement prediction results (predicted values ​​for specific survey items). The processor 101 that has obtained the predicted values ​​by another data complement prediction device 10' can also calculate predicted values ​​and corrected values ​​for specific survey items by itself, using the predicted values ​​as a reference.

[0019] The input device 104 can be configured with a keyboard, a touch panel, a microphone, buttons, etc. Using the input device 104, the user can input instructions (designation of target conditions and items to be predicted, which will be described later) and survey response results from previous surveys (time-series type 1 survey data and type 2 survey data).

[0020] The output device 105 can be configured with a display, speaker, printer, etc., and outputs the results of data complement prediction (predicted values ​​of survey items in a specific future Type 1 survey or survey items in a specific Type 2 survey).

[0021] (iii) The predicted value calculation unit 1011 constructed by the processor 101 calculates a predicted value for the future (e.g., next year) of a survey item (specified item) specified by the user, for example, from the aggregated values ​​of first-type survey data over a specified number of years in the past, using a specified method (e.g., a method that enables time series prediction, such as exponential smoothing).

[0022] The correlation calculation unit 1012 calculates the correlation (correlation coefficient or Euclidean distance value of time series characteristics) between the time series data of the above-mentioned specified item (for example, time series data extracted from past type 1 survey data of only the data of the specified item) and survey items other than the specified item, and performs processing to extract a predetermined number of items (for example, the top 10 items) that are highly correlated with the specified item.

[0023] The predicted value corrector 1013 calculates the correlation coefficients for the group of items highly related to the specified item acquired by the correlation calculator 1012. Large-scale language Model (For example, using GPT-4o) to obtain information on correction suggestions. Then, the predicted value correction unit 1013 converts the correction suggestions into correction values, and calculates the predicted value (after correction) by reflecting the correction values ​​on the predicted value calculated by the predicted value calculation unit 1011 (by multiplying, adding, or subtracting a correction coefficient).

[0024] <Outline of data complement prediction processing> 3 is a diagram for explaining an outline of the data interpolation prediction process according to this embodiment. The data interpolation prediction process does not simply execute a time-series prediction process on the aggregated values ​​of the specified items of the target data, but also calculates a correction value based on items other than the specified item included in the target data that have a high correlation with the specified item, and reflects the correction value in the predicted value obtained by the time-series prediction to calculate a highly accurate predicted value.

[0025] Time series forecasting methods such as exponential smoothing are processes that predict future states based on trends in past data. Therefore, if past data shows a linearly proportional (increasing or decreasing) trend, it is highly likely that the future will follow a similar trend and a forecast value can be calculated.

[0026] However, when unusual circumstances or environments arise, such as the COVID-19 pandemic, past trends do not necessarily continue into the future. For example, suppose you want to predict YouTube usage rates. For example, suppose past survey responses were 10% in 2014, 15% in 2015, 18% in 2016, 20% in 2017, 22% in 2018, and 23% in 2019. Given the changes in social conditions that have occurred since the COVID-19 pandemic in 2020, such as a sharp increase in homeworking and increased internet usage, simply forecasting future data based on past trends is unlikely to produce accurate predictions. Thus, numerical calculations alone cannot account for trends and changes in the world.

[0027] Therefore, in this embodiment, in parallel with the numerical calculation process (time series prediction process), Large-scale language Model of Use, By correcting the predicted values ​​obtained through numerical calculation processing, trends and changes in the world are reflected in the predicted values.

[0028] First, the user uses the input device 104 to specify a target condition indicating a time condition such as the year for which prediction is desired, and an item 301 for which prediction is desired. For example, the user specifies 2024 as the target condition and YouTube usage rate as the item for which prediction is desired (specified item). Note that, in addition to the time condition, it is also possible to set attribute conditions such as "males in their 20s" as the target condition.

[0029] The data complementation and prediction device 10 acquires time-series type 1 survey data (individual forms) 302 provided from the storage device 102 or externally. The data to be processed is not limited to type 1 survey data, but may be type 2 survey data, time-series survey response result data including multiple survey items, or any other type of time-series data including multiple items.

[0030] The data interpolation and prediction device 10 aggregates the acquired first-class survey data for each survey item, obtains an aggregate value 303 , and applies exponential smoothing to the aggregate value for the specified item to calculate a predicted value 304 .

[0031] In addition, in parallel with the calculation of the predicted value 304, or after (or before) calculating the predicted value 304, the data complement / prediction device 10 calculates a relationship value (a relationship value 305 between questions within the first type survey: for example, a correlation coefficient) between the item specified by the user and other survey items included in the first type survey.

[0032] Next, the data complement / prediction device 10 sorts the first-type intra-survey question relation values ​​305 in descending order, extracts the top 10 survey items, and sets them as a group of highly related items 306 .

[0033] Furthermore, the data complement prediction device 10 selects a group of highly related items 306 of GPT-4o307 Enter and correction suggestion information (information suggesting what corrections should be made): Large-scale language Model of Use (These are output as comments rather than numbers.)

[0034] The data interpolation prediction device 10 then digitizes the correction suggestion information and reflects it in the predicted value 304 (by multiplying it by a coefficient, adding or subtracting a constant, etc.) to generate a predicted value (after correction) 308.

[0035] <Details of data interpolation prediction process> FIG. 4 is a flowchart for explaining the details of the data complement prediction process according to this embodiment. FIG. 4 is a diagram showing an outline of the data complement prediction process described in FIG. 3 in the form of a flowchart. The main actors performing the operations in each step basically include a predicted value calculation unit 1011, a correlation calculation unit 1012, and a predicted value correction unit 1013. However, since these are processing units generated by deploying a data complement prediction program on the processor 101, they may be understood as the processor 101. Note that steps S401 to S403 (numerical calculation processing of predicted values) and steps S404 to S406 (processing for generating correction suggestion information) may be executed in reverse order or may be executed in parallel. FIG. 4 shows that the numerical calculation processing of predicted values ​​is executed first.

[0036] (i) Step 401 The processor 101 accepts the target condition (target time information such as the year to be predicted) and the predicted survey item (specified item) input from the input device 104. For example, the target condition can be 2024, and the specified item can be YouTube usage rate.

[0037] (ii) Step S402 The prediction value calculation unit 1011 acquires past (time-series) type 1 survey data (individual ballots) from the storage device 102 and calculates the aggregated value of the survey item (specified item) to be predicted in a time series (for example, for each year). This allows for the time-series trend of the numerical value (response result) of the specified item to be obtained. As mentioned above, the data to be subjected to prediction processing is not limited to the items of type 1 survey data, and may be any type of data as long as it is time-series data. For example, type 2 survey data, item-specific cabinet approval rating data, item-specific audience rating data, etc. may be used.

[0038] (iii) Step S403 The predicted value calculation unit 1011 executes time-series prediction processing on the aggregated values ​​of the specified items to calculate predicted values ​​for the target conditions (specified year) (numerical calculation processing). Here, exponential smoothing, ARIMA model, recurrent neural network, etc. can be used as the time-series prediction processing.

[0039] (iv) Step S404 The correlation calculation unit 1012 calculates a correlation value between each survey item (each question in the questionnaire) of the past type 1 survey data (individual sheets) acquired from the storage device 102 and the specified item. The correlation value can be calculated using, for example, a process for calculating Pearson's product-moment correlation coefficient, a process for calculating the distance (Euclidean distance) between the time series characteristics of the specified item and the time series characteristics of each survey item, or at least one of other correlation value calculation methods.

[0040] (v) Step S405 The correlation calculation unit 1012 extracts a predetermined number (for example, the top 10) of survey items with the highest numerical values ​​from the relationship values ​​calculated in step S404. The relationship values ​​may use correlation coefficients, distances of time-series characteristics, etc., but the top survey items may be extracted based on multiple indexes rather than just one index.

[0041] (vi) Step S406 The prediction correction unit 1013 extracts the highly correlated items extracted in step S405. On a large scale language Model (For example, GPT-4o) EnterThe system then generates information suggesting corrections to the predicted value (comments on what corrections should be made). Because the group of highly correlated items described above has the tendency to change depending on the state of the world, correcting the predicted value by taking into account information from these items makes it possible to reflect recent and current events and factors occurring in the world in the predicted value. For example, if the living environment is affected by the COVID-19 pandemic, people's behavior and lifestyles will have changed compared to before the pandemic (increased rate of people staying at home, increased internet usage, fewer opportunities to eat out, etc.), and these circumstances will also affect the trends of the specified items. Therefore, by reflecting such state of the world in the predicted value obtained by calculation, it becomes possible to obtain a predicted value that accurately matches the current situation.

[0042] (vii) Step S407 The prediction correction unit 1013 corrects the predicted value (predicted value obtained by numerical calculation processing) calculated in step S403 based on the correction suggestion information (reason for correction) obtained in step S406, and outputs the corrected value to the output device 105.

[0043] <Specific examples of forecast value correction> FIG. 5 is a table showing an example of the results of corrected predicted values ​​obtained by the data interpolation prediction process of this embodiment. The predicted value correction result table (example) 500 is a table showing various pieces of information obtained by correcting the predicted values ​​obtained from the aggregated values ​​of the first-class survey data, and includes the target conditions (2023) and items to be predicted 501, which correspond to reference number 301 in Figure 3, actual measured values ​​for a specified number of years in the past 502, and predicted values ​​for 2023 503 calculated based on exponential smoothing from the data from 2016 to 2022 of the actual measured values ​​502. Large-scale language Model The configuration items include a correction value 504 for 2023 based on the correction suggestion (corresponding to reason 505) obtained by (GPT-4o), a reason (correction suggestion) 505, and related information 506 corresponding to the highly related item group 306 in Figure 3.

[0044] In the predicted value correction result table (example) 500 , 2023 (actual measurement value) included in the actual measurement value 502 is information for verifying 2023 (predicted value) 503 .

[0045] For example, consider the case where the percentage of "frequently take domestic cruises / overseas cruises" in 2023 is predicted as the item 501 to be predicted. In this case, when the actual measured value 502 of "percentage of people who have taken domestic cruises / overseas cruises" from 2016 to 2022 is used to calculate the predicted value 503 for 2023 based on the exponential smoothing method, 0.0% is obtained. Meanwhile, when question items highly correlated with the question item "frequently take domestic cruises / overseas cruises" are extracted from the Type 1 survey data, the items shown in related information 506 are obtained. Then, the extracted related information 506 (response results for a total of 10 items: "gender and age composition ratio," "family composition," "annual household income," "would like to take domestic cruises / overseas cruises in the future (again)," "travel frequency," "frequently go swimming in the sea," "frequently go yachting / sailing," etc.) On a large scale language Model (GPT-4o) Enter As a result, a correction suggestion was obtained, as shown in reason 505: "Past related data shows signs of a gradual recovery in awareness of domestic / overseas cruises, and a certain level of interest was shown in particular in the 'Leisure / hobby activities that people would like to try (or continue to try) in the future' category." Therefore, based on this correction suggestion, the predicted value correction unit 1013 corrected the predicted value for 2023 from 0.0% to 1.0%. The correction of the predicted value can be obtained, for example, by calculating the predicted value for 2023 for each item (predicted value obtained by numerical calculation processing) from the trends of each item listed in the related information, and then taking their average or weighted average (for example, the weighting coefficient is set to a larger value for an item with a higher correlation coefficient).

[0046] <Summary> (i) This embodiment proposes a data prediction system (data complement prediction device 10) that predicts future data based on time series data (e.g., first-class survey data) acquired over a predetermined period (e.g., over the past few years). The data complement prediction device 10 uses time series data (time series trends of the numerical values ​​of the designated items obtained from past data) of desired items (survey items designated by the user: designated items) to calculate predicted values ​​for the desired items at a desired time period (such as a future time period (next year) when no survey has been conducted or a blank time period (year) in the past when no survey has been conducted) through numerical calculation processing. Here, exponential smoothing can be used as the numerical calculation processing. Furthermore, the data complement prediction device 10 extracts a predetermined number of items (other survey items in the first-class survey) having time series data that are highly correlated with the time series data of the designated items as highly correlated items. Next, the data complement prediction device 10 calculates the predicted values ​​for the extracted highly correlated items. On a large scale language Input to the model It generates suggested information for correction of the predicted value. Furthermore, it is possible to perform data supplementation prediction. Device The unit 10 corrects the forecast value obtained by the numerical calculation process based on the correction suggestion information and outputs the corrected forecast value. By doing so, it is possible to reflect in the forecast value the variable factors such as changes in the world environment and trends at the time of prediction execution, and it is possible to provide a highly accurate time series forecast value that matches the actual situation in the world.

[0047] (ii) The highly correlated items can be determined by calculating the correlation coefficient and / or distance value (Euclidean distance in the time series transition) between the time series data of the specified item and the time series data of multiple items included in the Type 1 survey data, and extracting a predetermined number of items (for example, the top 10) in descending order of correlation coefficient and / or a predetermined number of items (for example, the top 10) in descending order of distance value. By performing such correlation calculations, it is possible to extract survey items that reflect the actual state of the world.

[0048] (iii) Large-scale language ModelThe correction suggestion information obtained by (for example, GPT-4o) is a comment rather than a numerical value. Therefore, the prediction correction unit 1013 corrects the predicted value (for example, 2023 (predicted value) 503 in FIG. 5) obtained by numerical calculation processing (for example, exponential smoothing) so as to be consistent with the comment. This allows the predicted value to be corrected taking into account the trends of highly related items, making it possible to provide highly accurate time series predicted values.

[0049] (iv) The functions of the present embodiment can also be realized by software program code. In this case, a storage medium on which the program code is recorded is provided to a system or device, and the computer (or CPU or MPU) of the system or device reads the program code stored in the storage medium. In this case, the program code read from the storage medium itself realizes the functions of the above-mentioned embodiment, and the program code itself and the storage medium on which it is stored constitute the present disclosure. Examples of storage media for providing such program code include flexible disks, CD-ROMs, DVD-ROMs, hard disks, optical disks, magneto-optical disks, CD-Rs, magnetic tapes, non-volatile memory cards, and ROMs.

[0050] In addition, an operating system (OS) running on a computer may perform some or all of the actual processing based on instructions in the program code, and the functions of the above-described embodiments may be realized by this processing. Furthermore, after the program code is read from a storage medium and written to a memory on a computer, a CPU of the computer may perform some or all of the actual processing based on instructions in the program code, and the functions of the above-described embodiments may be realized by this processing.

[0051] Furthermore, the program code of the software that realizes the functions of the embodiments and each example may be distributed via a network and stored in a storage means such as a hard disk or memory of the system or device, or in a storage medium such as a CD-RW or CD-R, so that when in use, the computer (or CPU or MPU) of the system or device reads and executes the program code stored in the storage means or storage medium.

[0052] The processes and techniques described herein are not inherently related to any specific device and can be implemented by a combination of components. Various types of general-purpose devices can also be added. A dedicated device may be constructed to perform the functions of this embodiment and each example. Various functions can also be formed by appropriately combining multiple components disclosed in this embodiment and each example. For example, some components may be omitted from all the components shown in the embodiment and each example, or components from different examples may be appropriately combined.

[0053] Although specific embodiments are described in this disclosure, they are intended in all respects to be illustrative and not restrictive. Those skilled in the art will recognize that there are numerous combinations of hardware, software, and firmware suitable for implementing the disclosed technology. For example, the described software can be implemented in a wide variety of programming or scripting languages, such as assembler, C / C++, Perl, Shell, PHP, Java, etc.

[0054] Furthermore, in the above-described embodiment, the control lines and information lines are those that are considered necessary for the explanation, and not all control lines and information lines in the product are necessarily shown. All components may be interconnected.

[0055] In addition, other implementations of the present disclosure will be apparent to those skilled in the art from consideration of the present embodiments and examples. The specification and examples are exemplary only, with the scope and spirit of the technology of the present disclosure being indicated by the following claims. [Explanation of symbols]

[0056] 10 Data Complement Prediction Device (System) 101 Processor (control device) 102 Storage Devices 103 Communication Devices 104 Input Devices 105 Output Devices 1011 Prediction value calculation unit 1012 Correlation calculation unit 1013 Prediction value correction unit

Claims

1. A data prediction system that predicts future data based on time series data acquired over a predetermined period, a storage device that stores a program for performing time series prediction on a desired item among a plurality of items based on an aggregate value of time series data on the plurality of items; a processor that reads the program from the storage device and executes it; The processor: A process of calculating a predicted value of the desired item at a desired time using time series data of the desired item; a process of extracting a predetermined number of items having time series data highly correlated with the time series data of the desired item from the plurality of items as highly correlated items; a process of inputting the predetermined number of highly correlated items into a large-scale language model to generate suggestion information for correcting the predicted value; a process of correcting the predicted value based on the correction suggestion information and outputting a corrected predicted value; A data prediction system that performs

2. In claim 1, The processor calculates a correlation coefficient and / or a distance value between the time series data of the desired item and the time series data of the plurality of items, and extracts the predetermined number of items in descending order of the correlation coefficient and / or the predetermined number of items in descending order of the distance value as the highly correlated items.

3. In claim 1, The processor digitizes the correction suggestion information and corrects the predicted value.

4. In claim 1, The processor processes time series data for each item of the individual data including the plurality of items.

5. In claim 4, A data prediction system, wherein the individual data is survey data on a plurality of items related to people's lives.

6. In claim 1, The processor calculates the predicted value by applying exponential smoothing to time series data of the desired item.

7. A data prediction method for predicting future data based on time series data acquired over a predetermined period, comprising: a processor reading from a storage device a program for time-series prediction of a desired item among a plurality of items based on an aggregate value of time-series data for the plurality of items; The processor calculates a predicted value of the desired item at a desired time using time series data of the desired item; The processor extracts a predetermined number of items having time series data highly correlated with the time series data of the desired item from the plurality of items as highly correlated items; the processor inputs the predetermined number of highly correlated items into a large-scale language model to generate suggested information for correction to the predicted value; the processor corrects the predicted value based on the correction suggestion information and outputs the corrected predicted value; A data prediction method comprising:

Images