JP7665697B2 - Information providing device, information providing method, and information providing program - Google Patents

Description

The present invention relates to an information providing device, an information providing method, and an information providing program.

When advancing an analysis project, it is important to combine the analyst's analytical skills with data-specific domain knowledge, and close communication between the analyst and project stakeholders is essential to moving the project forward.

In recent years, data analysis tools are increasingly being provided in so-called no-code data analysis tool (e.g., application) environments. No-code data analysis tools have the advantage that they do not require programming skills and can be used intuitively even by beginners. This has partially reduced the coding and other labor required for analysis.

JP 2023-27964 A

“No-code AI development tool Node-AI”, [online], [Retrieved July 21, 2023], Internet <https://sdpf.ntt.com/services/nodeai/>

When an analysis project is carried out using a data analysis tool, it is necessary to share information such as the analysis results among project participants. However, there is no tool that allows discussions about the analysis to be concluded within the tool itself. In addition, project personnel (hereafter referred to as stakeholders) other than the person in charge of analysis (hereafter referred to as the analyst) have difficulty understanding the content of the analysis just by looking at the analysis tool screen.

For this reason, analysts prepare separate reports based on the results of their analysis using the tools. However, analysts are responsible for many tasks, such as deciding the analysis policy, conducting the analysis, and preparing the reports, and the task of preparing separate reports, especially for reports to other stakeholders, places a heavy burden on analysts.

The present invention has been made in consideration of the above, and aims to provide an information provision device, an information provision method, and an information provision program that can reduce the burden of report creation work on analysts.

In order to solve the above-mentioned problems and achieve the object, the information providing device of the present invention is characterized by having a prompt creation unit that creates a prompt for generating a sentence showing an overview of a series of processes in a language model based on execution parameters including parameters of each process and execution results of each process received from an analysis device that executes a data analysis tool that generates a series of processes related to data analysis, a summary generation unit that generates a sentence showing an overview of the series of processes by inputting the prompt generated by the prompt creation unit into the language model, and an expression conversion unit that converts the sentence showing the overview generated by the summary generation unit into an expression suitable for a report and displays the converted sentence showing the overview of the series of processes on a terminal device.

The present invention can reduce the burden on analysts in creating reports.

FIG. 1 is a diagram showing an example of a configuration of an information providing system according to the first embodiment. FIG. 2 is a diagram showing an example of canvas data. FIG. 3 is a diagram showing an example of a summary list displayed on the screen of a related party terminal. FIG. 4 is a diagram illustrating a process flow of the information providing system in the first embodiment. FIG. 5 is a sequence diagram showing a processing procedure of the information providing process according to the first embodiment. FIG. 6 is a diagram illustrating an example of a configuration of an information providing system according to the second embodiment. FIG. 7 is a diagram illustrating a process flow of the information providing system in the second embodiment. FIG. 8 is a sequence diagram showing a processing procedure of the information providing process according to the second embodiment. FIG. 9 is a diagram illustrating an example of a computer that realizes each of the components of the information providing system by executing a program.

One embodiment of the present invention will be described in detail below with reference to the drawings. Note that the present invention is not limited to this embodiment. In addition, in the drawings, the same parts are denoted by the same reference numerals.

[First embodiment]
[Configuration of First Embodiment]
The configuration of the information providing system will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of the configuration of the information providing system according to the first embodiment. As shown in Fig. 1, the information providing system 100 has an analyst terminal 20, an analysis server 30 (analysis device), an information providing device 10, and a related party terminal 40 (terminal device). The information providing device 10 is a server connected to the analysis server 30 via a network. The analyst terminal 20 is connected to the analysis server 30 via a network. The related party terminal 40 is connected to the information providing device 10 via a network.

The analyst terminal 20 is a terminal device used by an analyst of a data analysis project. The analyst terminal 20 is a PC (Personal Computer), a notebook PC, a tablet terminal, a smartphone, or the like. The analyst sets the analysis policy and carries out the analysis. Specifically, the analyst is responsible for setting a series of processes related to data analysis to be executed by the analysis server 30. Under the analyst's operation, the analyst terminal 20 communicates with the analysis server 30, sets a series of processes related to data analysis (including setting parameters for each process), and causes the analysis server 30 to execute the processes related to data analysis.

The analysis server 30 has the functionality of a data analysis tool such as Node-AI described in Non-Patent Document 1. The analysis server 30 may have functionality equivalent to the information processing device described in Patent Document 1. The analyst executes the data analysis tool of the analysis server 30 via the analyst terminal 20 to generate canvas data.

The data analysis tool generates a series of processes (pipelines) for data analysis using each machine learning model in response to the analyst's operations. Each machine learning model that performs data analysis is called an analytical model. Examples of processes include preprocessing of the data, learning of the analytical model using the data (updating parameters), evaluation of the trained analytical model, etc. Note that the division of processes in the data analysis tool is not limited to preprocessing, learning, and evaluation, and may be further subdivided.

Node-AI employs flow-based programming, which allows designing a processing pipeline for data without coding. Node-AI also allows intuitive description of a processing pipeline for data by connecting the data to be processed with the functions that can execute each process. In this first embodiment, a function that executes a specific process on data is referred to as a "card." Furthermore, data for which a pipeline is described is referred to as "canvas data."

Figure 2 is a diagram showing an example of canvas data. In the example shown in Figure 2, the canvas data includes multiple cards N1, N2, N3, N41, N42, N51, N52, N6, N7, N8, and N9, which are interconnected. Cards N1, N2, N3, N41, N42, N51, N52, N6, N7, N8, and N9 correspond to functions that execute specified data processing. The information of the interconnected cards N1, N2, N3, N41, N42, N51, N52, N6, N7, N8, and N9 corresponds to a "graph."

Card N1 is a card that specifies the data to be processed (e.g., time series data).

Cards N2, N3, N41, N42, N51, and N52 correspond to processes that further subdivide the pre-processing of data. Each of cards N2, N3, N41, N42, N51, and N52 takes the processing results of the previous card as input, executes its respective function, and outputs the processing results. Card N2 executes missing value interpolation, card N3 executes data division, cards N41 and N42 execute normalization, and cards N51 and N52 execute time window extraction.

Cards N6 and N7 correspond to the process of learning an analytical model using data. Card N6 executes a linear model, and card N7 executes learning (parameter updates).

Cards N8 and N9 correspond to the process of evaluating the trained analytical model. Card N8 executes the evaluation, and card N9 performs the factor analysis.

Note that the canvas data is not limited to the example described in FIG. 2 and may also have a card that performs other data processing.

Then, in the analysis server 30, the data analysis tool outputs the execution parameters when executing the series of processes to the information providing device 10. The execution parameters are information including the parameters of each process in the series of processes and the execution results of each process.

The information providing device 10 generates a document that shows an overview of a series of processes related to the data analysis based on the execution parameters received from the analysis server 30. The overview summarizes what processes were performed in this analysis.

At this time, the information providing device 10 converts the text outlining the series of processes into an expression suitable for a report, and displays the text outlining the series of processes converted into an expression suitable for a report on the screen of the relevant person's terminal 40.

Specifically, the information providing device 10 generates an overview list in which each sentence indicating the overview of the process, converted into an expression suitable for a report, is associated with the identification information of each process. The information providing device 10 generates an overview list of the report format based on the execution parameters received from the analysis server 30, and displays it on the relevant party terminal 40.

The stakeholder terminal 40 is a terminal device used by the stakeholders in the data analysis project. The stakeholder terminal 40 is a PC, a notebook PC, a tablet terminal, a smartphone, etc.

The relevant party terminal 40 communicates with the analysis server 30 based on the operation of the relevant party, and displays and outputs the summary list sent from the analysis server 30. The summary list is converted into an expression suitable for a report. The relevant party can understand the analysis contents by visually checking the summary list displayed on the relevant party terminal 40.

[Information Providing Device]
Next, returning to FIG. 1, the information providing device 10 will be described.

The information providing device 10 is realized, for example, by loading a specific program into a computer or the like including a ROM (Read Only Memory), a RAM (Random Access Memory), a CPU (Central Processing Unit), etc., and having the CPU execute the specific program. The information providing device 10 also has a communication interface for transmitting and receiving various types of information to and from other devices (for example, the analysis server 30, the stakeholder terminal 40) connected via a network or the like.

The information providing device 10 has a prompt creation unit 11, an overview generation unit 12, and an expression conversion unit 13.

The prompt creation unit 11 creates a prompt for causing the language model of the summary generation unit 12 (described later) to generate a sentence showing an overview of the process based on the execution parameters received from the analysis server 30, including the parameters of each process and the execution results of each process. The prompt becomes the input of the language model. The prompt creation unit 11 outputs the generated prompt to the summary generation unit 12. For example, the prompt creation unit 11 converts data including the execution parameters into text with a consistent format so that it can be processed by the language model of the summary generation unit 12 (described later).

The summary generation unit 12 generates a sentence that shows an overview of a series of processes related to data analysis by inputting the prompt generated by the prompt creation unit 11 into a language model. The summary generation unit 12 outputs the generated sentence that shows an overview of the series of processes to the expression conversion unit 13.

The summary generation unit 12 has multiple language models. A language model is provided for each process related to data analysis. Each language model has been trained in advance to generate sentences that show a summary of the process according to the execution parameters of the corresponding process.

Each language model in the summary generation unit 12 is a large language model (LLM). Each language model in the summary generation unit 12 may be a neural network. Each language model in the summary generation unit 12 may be a language model group including multiple language models.

Each language model in the summary generation unit 12 performs inference based on the input prompt and generates a sentence in natural language obtained by the inference. Each language model generates a sentence that shows an overview of the corresponding process. The sentences generated by each language model are colloquial sentences.

An example will be described in which the data analysis processes are preprocessing of the data, learning an analytical model using the data, and evaluating the learned analytical model. In this case, the summary generation unit 12 includes a language model whose parameters are tuned to suit each of these three processes.

The summary generation unit 12 has a preprocessing model 121, which is a language model whose parameters are tuned for preprocessing, a learning model 122, which is a language model whose parameters are tuned for learning processing, and an evaluation model 123, which is a language model whose parameters are tuned for evaluation processing.

Specifically, the preprocessing model 121 is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the preprocessing and a sentence that shows an overview of the preprocessing created by an experienced analyst in response to the execution parameters. The preprocessing model 121 can appropriately generate a sentence that shows an overview of the preprocessing that corresponds to the execution parameters.

Similarly, the learning model 122 is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the learning process and a sentence that shows an overview of the learning process created by a veteran analyst in response to the execution parameters. The evaluator model 123 is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the evaluation process and a sentence that shows an overview of the evaluation process created by a veteran analyst in response to the execution parameters.

The division of the processes of the data analysis tool is not limited to the above, and may be further subdivided into pre-processing, learning, and evaluation. Accordingly, the summary generation unit 12 has a language model tuned to each of the subdivided processes. For example, the summary generation unit 12 includes a language model tuned to each of the cards in the data analysis tool. In the example of FIG. 2, the summary generation unit 12 has a missing value interpolation model tuned to the missing value interpolation process of card N2, a data partitioning model tuned to the data partitioning process of card N3, and the like.

The expression conversion unit 13 converts the text indicating the summary generated by the summary generation unit 12 into an expression suitable for a report. The expression conversion unit 13 has a report material manager model 131.

The report material manager model 131 receives as input multiple summary sentences generated by each language model (e.g., preprocessing model 121, learning model 122, evaluation model 123) and converts them into expressions suitable for a report. The report material manager model 131 then outputs the sentences that summarize each process, converted into expressions suitable for a report, as a single piece of data associated with the identification information of each process. The data output by the report material manager model 131 is explained as a list of summaries.

The expression conversion unit 13 inputs each sentence that indicates an overview of a process generated by each language model into the report material manager model 131, and generates an overview list in which each sentence is converted into an expression suitable for a report and associated with the identification information of each process.

The expression conversion unit 13 displays a text (summary list) outlining the series of processes, converted into an expression suitable for a report, on the relevant party terminal 40.

Figure 3 is a diagram showing an example of a summary list displayed on the screen of the stakeholder terminal 40. As shown in Figure 3, in the summary list, a sentence indicating the purpose of the analysis is shown as a heading (box B10). In addition, in the summary list, the time series data to be processed (card N1) is shown in text and table format in box B11.

Furthermore, in the summary list, each card N2, N3, N41, N42, N51, N52, and N8 of the canvas data shown in Figure 2 is displayed in association with a sentence (boxes B2, B3, B4, B5, and B8) indicating the corresponding process and its summary.

For example, for card N2, text C2 is displayed in box B2 to the right of card N2 indicating that the process corresponding to card N2 is "missing value interpolation" and stating, "Linear interpolation was performed by referencing the values before and after where missing values existed in the data. Interpolation was performed for four missing values."

For card N8, in box B8 to the left of card N8, it is displayed that the process corresponding to card N8 is "evaluation," and as an overview of "evaluation" (explanation of the evaluation indicators and evaluation results), text C8 stating "PMSE and MAE refer to the error between predicted values and actual values, and the smaller the error, the higher the prediction accuracy. RMSE takes a large value when the evaluation data contains outliers. Looking at RMSE, the constructed model will have an error of 6974 on average between predicted values and actual values." and a list table T8 of the evaluation results are displayed.

For the other cards N3, N41, N42, N51, and N52, text C3, C4, and C5 indicating the corresponding process and its overview are displayed to the right of each card N3, N41, N42, N51, and N52 (frames B3, B4, and B5).

And each of the texts C2, C3, C4, C5, and C8 is written in a way that is suitable for a report.

[Process flow]
Next, a process flow of the information providing system in the embodiment 1 will be described. Fig. 4 is a diagram for explaining a process flow of the information providing system in the embodiment 1. Fig. 5 is a sequence diagram showing a process procedure of the information providing process according to the embodiment 1.

As shown in FIG. 4, the analysis server 30 executes an analysis process ((A) in FIG. 4). Specifically, the analyst operates the analyst's terminal 20 to send an analysis request to the analysis server 30 (step S1 in FIG. 5). In response to an instruction from the analyst terminal 20, the analysis server 30 sets a series of processes related to data analysis and executes the processes related to data analysis ((1) in FIG. 4, step S2 in FIG. 5).

The analyst terminal 20 adjusts the parameters of each process through the analyst's operation (step S3 in FIG. 5). If the analyst has not completed adjusting the parameters of each process (step S4 in FIG. 5: No), the analysis server 30 returns to executing the analysis process for readjustment.

When the analyst has completed adjusting the parameters for each process (step S4 in FIG. 5: Yes), the analysis server 30 passes the execution parameters (meaning the parameters and execution results for each process) to the information providing device 10 ((2) in FIG. 4, step S5 in FIG. 5).

The information providing device 10 creates an overview of a series of processes related to data analysis (Figure 4 (B)).

Specifically, in the information providing device 10, the prompt creation unit 11 creates a prompt based on the execution parameters received from the analysis server 30 ((3) in FIG. 4, step S6 in FIG. 5). The prompt becomes input to each language model of the summary generation unit 12 (e.g., the preprocessing model 121, the learning model 122, and the evaluation model 123).

Next, in the information providing device 10, the summary generating unit 12 generates a sentence outlining a series of processes related to data analysis by inputting the prompt generated by the prompt creating unit 11 into each language model (e.g., the preprocessing model 121, the learning model 122, and the evaluation model 123) (step S7 in FIG. 5). A prompt is input for each language model corresponding to each process, and each language model outputs a colloquial summary of the process ((4) in FIG. 4).

In the information providing device 10, the expression conversion unit 13 uses the report material manager model 131 to convert the sentences indicating the summaries generated by the summary generation unit 12 into expressions suitable for reports ((5) in FIG. 4, step S8 in FIG. 5). The colloquial process summaries generated by each language model are input to the report material manager model 131. The report material manager model 131 converts each colloquial process summary into a report-like feel, that is, into expressions suitable for reports. The report material manager model 131 outputs a list of summaries in which the sentences indicating the summaries of each process converted into expressions suitable for reports are associated with the identification information of each process.

The information providing device 10 transmits the summary list output by the report material manager model 131 to the stakeholder terminal 40 (step S9 in FIG. 5) and displays it on the screen (tool) of the stakeholder terminal 40 ((6) in FIG. 4, step S10 in FIG. 5).

[Effects of the First Embodiment]
In this way, the information providing device 10 according to the first embodiment automatically generates an outline list of a series of processes related to data analysis, expressed in a format suitable for a report, based on the execution parameters received from the analysis server 30, and displays it on the related party terminal 40. According to the first embodiment, it is possible to support the related party in understanding the analysis content while reducing the burden on the analyst in creating the report.

[Embodiment 2]
Next, a description will be given of embodiment 2. In the information providing system according to embodiment 2, a summary list is generated that reflects the analyst's aims and/or ideas for a series of processes related to data analysis.

[Configuration of the Second Embodiment]
The configuration of the information providing system will be described with reference to Fig. 6. Fig. 6 is a diagram showing an example of the configuration of the information providing system according to the embodiment 2. The information providing system 200 shown in Fig. 6 has an information providing device 210 instead of the information providing device 10 shown in Fig. 1.

The information providing device 210 generates an overview list of a series of processes related to data analysis, expressed in a format suitable for a report. At this time, the information providing device 210 creates a prompt and generates the overview list based on the analyst's aims and/or ideas for the series of processes related to data analysis, in addition to the execution parameters received from the analysis server 30.

[Information Providing Device]
An information providing apparatus 210 shown in Fig. 6 will be described. The information providing apparatus 210 has a prompt creating section 211 and a summary creating section 212 instead of the prompt creating section 11 and the summary creating section 12 shown in Fig. 1. The information providing apparatus 210 further has an extraction section 214.

The extraction unit 214 uses a dialogue model (machine learning model) to engage in a dialogue with the analyst via the analyst terminal 20. The extraction unit 214 extracts the analyst's aims and/or ingenuity for a series of processes related to data analysis from the dialogue response between the analyst and the dialogue model. The extraction unit 214 has a conversation manager model 2141, a dialogue summary model 2142, and a state detection model 2143 as dialogue models.

The conversation manager model 2141 integrates the inputs of the conversation summary model 2142 and the state detection model 2143 to generate a coherent and comprehensive conversation. The conversation manager model 2141 extracts the analyst's aims and/or ideas for a series of processes related to data analysis from the conversation responses between the analyst and the conversation model.

The dialogue summary model 2142 takes as input the analyst's responses to the questions generated by the dialogue manager model 2141 and determines the dialogue summary. The dialogue summary model 2142 generates a concise summary of the ongoing conversation. The dialogue summary model 2142 considers the most recent analyst utterance, past conversation history, and current responses from the analyst to generate a dialogue summary that maintains consistency and context throughout the entire conversation.

The state detection model 2143 detects the state of the dialogue based on the dialogue summary by the dialogue summary model 2142. The state detection model 2143 determines whether the analyst should move to the next state in the dialogue based on the user's needs, the accuracy of the objective, the smooth flow of the conversation, and the decision on whether to proceed to the next dialogue state.

The conversation manager model 2141 determines whether an answer to the question has been obtained based on the detection results by the state detection model 2143.

If no answer to the question is obtained, the conversation manager model 2141 generates a sentence that takes into consideration the overall flow of the conversation and outputs it as a question to the analyst. If an answer to the question is obtained, the conversation manager model 2141 inputs the extracted analyst's aim and/or ingenuity for the series of processes related to data analysis to the prompt creation unit 211 in order to incorporate the analyst's aim and/or ingenuity for the series of processes related to data analysis into the prompt.

The prompt creation unit 211 has a prompt creation model 2111. The prompt creation unit 211 inputs the execution parameters received from the analysis server 30 and the analyst's aim and/or ingenuity extracted by the extraction unit 214 to the prompt creation model 2111, thereby creating a prompt for causing the language model of the summary generation unit 212 (described later) to generate a sentence showing an overview of the process.

The prompt creation model 2111 is a machine learning model in which machine learning is performed using learning data so as to input the analyst's goals and/or ideas and the execution parameters of the analysis process, and output a prompt corresponding to the analysis process.

Similar to the summary generation unit 12, the summary generation unit 212 generates a sentence that shows an overview of a series of processes related to data analysis by inputting the prompt generated by the prompt creation unit 211 into multiple language models (e.g., a preprocessing model 2121, a learning model 2122, and an evaluation model 2123).

A language model is provided for each process related to data analysis. Each language model is pre-trained to generate sentences that outline the process according to the execution parameters of the corresponding process and the analyst's aims and/or ideas.

Specifically, the preprocessing model 2121 is a machine learning model in which machine learning is performed using as learning data the execution parameters of the preprocessing, prompts based on the analyst's aims and/or ingenuity for the preprocessing, and sentences outlining the preprocessing created by an experienced analyst corresponding to the execution parameters and the analyst's aims and/or ingenuity for the preprocessing. The preprocessing model 2121 reflects the analyst's aims and/or ingenuity for the preprocessing and can appropriately generate sentences outlining the preprocessing corresponding to the execution parameters.

Similarly, the learning model 2122 is a machine learning model in which machine learning is performed using as learning data the execution parameters of the learning process, prompts based on the analyst's aims and/or ingenuity for the learning process, and text outlining the learning process created by an experienced analyst in response to the execution parameters and the analyst's aims and/or ingenuity for the learning process.

The evaluator model 2123 is a machine learning model in which machine learning is performed using as learning data the execution parameters of the evaluation process, prompts based on the analyst's aims and/or ingenuity for the evaluation process, and text outlining the evaluation process created by a veteran analyst in response to the execution parameters and the analyst's aims and/or ingenuity for the evaluation process.

[Process flow]
Next, a process flow of the information providing system in the embodiment 2 will be described. Fig. 7 is a diagram for explaining a process flow of the information providing system in the embodiment 2. Fig. 8 is a sequence diagram showing a process procedure of the information providing process according to the embodiment 2.

As in the first embodiment, the analysis server 30 executes the analysis process ((A) in FIG. 7). The analyst terminal 20 transmits an analysis request to the analysis server 30 (step S21 in FIG. 8). In response to an instruction from the analyst terminal 20, the analysis server 30 sets a series of processes related to data analysis and executes the processes related to data analysis ((1) in FIG. 7, step S22 in FIG. 8).

The analyst terminal 20 adjusts the parameters of each process through the analyst's operation (step S23 in FIG. 8). If the analyst has not completed adjusting the parameters of each process (step S24 in FIG. 8: No), the analysis server 30 returns to executing the analysis process for readjustment.

When the analyst has completed adjusting the parameters for each process (step S24 in FIG. 8: Yes), the analysis server 30 passes the execution parameters (meaning the parameters and execution results for each process) to the information providing device 210 ((2) in FIG. 7, step S25 in FIG. 8).

In the information providing device 210, the extraction unit 214 uses the dialogue model to dialogue with the analyst via the analyst terminal 20, thereby extracting the analyst's aims and/or ingenuity for each process in a series of processes related to data analysis. Specifically, in the information providing device 210, the conversation manager model 2141 generates sentences that take into account the overall flow of the conversation, outputs these to the analyst terminal 20 as questions to the analyst, and receives answers to these questions from the analyst terminal 20, thereby generating a dialogue with the analyst (step S26 in FIG. 8).

In the extraction unit 214, the dialogue summary model 2142 receives as input the analyst's response to the question generated by the dialogue manager model 2141 ((3-1) in FIG. 7) and determines the summary of the dialogue ((3-2) in FIG. 7, step S27 in FIG. 8).

In the extraction unit 214, the state detection model 2143 detects the state of the dialogue with the analyst based on the dialogue summary by the dialogue summary model 2142 ((3-2) in FIG. 7, step S28 in FIG. 8). The conversation manager model 2141 determines whether or not an answer to the question has been obtained based on the detection result by the state detection model 2143 ((3-3) in FIG. 7, step S29 in FIG. 8).

If the question cannot be answered because there are ambiguous parts ((3-4) in Figure 7, step S29 in Figure 8: No), the process returns to step S27 in Figure 8, and the conversation manager model 2141 generates a sentence that takes into account the overall flow of the conversation and outputs it as a question to the analyst ((3-4) in Figure 7).

If an answer to the question is obtained (step S29 in FIG. 8: Yes), the extraction unit 214 extracts the analyst's aim and/or ingenuity for the series of processes related to data analysis. The extraction unit 214 then inputs the extracted aim and/or ingenuity for the series of processes related to data analysis to the prompt creation unit 211 (step S30 in FIG. 8), and incorporates the analyst's aim and/or ingenuity for the series of processes related to data analysis into the prompt.

The prompt creation unit 211 creates a prompt by inputting the execution parameters received from the analysis server 30 and the analyst's aim and/or ingenuity points extracted by the extraction unit 214 to the prompt creation model 2111 ((4) in FIG. 7), step S31 in FIG. 8). The prompt creation model 2111 receives the execution parameters and the analyst's aim and/or ingenuity points received from the analysis server 30 as input, and generates and outputs input text (prompt) for each language model based on the execution parameters and the analyst's aim.

Next, in the information providing device 210, the summary generating unit 212 inputs the prompt generated by the prompt creating unit 211 into each language model (e.g., the preprocessing model 2121, the learning model 2122, and the evaluation model 2123) to generate a sentence outlining a series of processes related to data analysis (step S32 in FIG. 8). A prompt is input for each language model corresponding to each analysis process, and each language model outputs a colloquial summary of the process ((5) in FIG. 7). Each of these process summaries corresponds to the execution parameters and the analyst's aims and/or elaboration for the evaluation process.

In the information providing device 210, the expression conversion unit 13 converts the sentences indicating the summaries generated by the summary generation unit 212 into expressions suitable for reports using the report material manager model 131 ((6) in FIG. 7, step S33 in FIG. 8). The expression conversion unit 13 creates a list of summaries in which the sentences indicating the summaries of each process, converted into expressions suitable for reports, are associated with the identification information of each process.
The information is transmitted to the terminal 40 of the person concerned (step S34 in FIG. 8) and displayed on the screen (tool) of the terminal 40 of the person concerned ((7) in FIG. 7, step S35 in FIG. 8).

[Effects of the second embodiment]
The information providing device 210 according to the second embodiment can generate a summary list that reflects the analyst's aims and/or ideas for a series of processes related to data analysis, and can facilitate assistance to relevant parties in understanding the analysis content.

[System Configuration of the Embodiment]
Each component of the information providing systems 100 and 200 is a functional concept, and does not necessarily have to be physically configured as shown in the drawings. In other words, the specific form of distribution and integration of the functions of the information providing systems 100 and 200 is not limited to that shown in the drawings, and all or part of them can be functionally or physically distributed or integrated in any unit depending on various loads, usage conditions, etc.

Furthermore, each process performed in the information provision systems 100 and 200 may be realized, in whole or in part, by a CPU, a GPU (Graphics Processing Unit), and a program analyzed and executed by the CPU and GPU. Furthermore, each process performed in the information provision systems 100 and 200 may be realized as hardware using wired logic.

Furthermore, among the processes described in the embodiments, all or part of the processes described as being performed automatically can be performed manually. Alternatively, all or part of the processes described as being performed manually can be performed automatically using a known method. In addition, the information including the processing procedures, control procedures, specific names, various data, and parameters described above and illustrated in the drawings can be changed as appropriate unless otherwise specified.

[program]
9 is a diagram showing an example of a computer in which each component device of the information provision systems 100 and 200 is realized by executing a program. The computer 1000 has, for example, a memory 1010 and a CPU 1020. The computer 1000 also has a hard disk drive interface 1030, a disk drive interface 1040, a serial port interface 1050, a video adapter 1060, and a network interface 1070. Each of these components is connected by a bus 1080.

The memory 1010 includes a ROM 1011 and a RAM 1012. The ROM 1011 stores a boot program such as a BIOS (Basic Input Output System). The hard disk drive interface 1030 is connected to a hard disk drive 1090. The disk drive interface 1040 is connected to a disk drive 1100. A removable storage medium such as a magnetic disk or optical disk is inserted into the disk drive 1100. The serial port interface 1050 is connected to a mouse 1110 and a keyboard 1120, for example. The video adapter 1060 is connected to a display 1130, for example.

The hard disk drive 1090 stores, for example, an OS (Operating System) 1091, an application program 1092, a program module 1093, and program data 1094. That is, the programs that define each process of the information provision systems 100 and 200 are implemented as program modules 1093 in which code executable by the computer 1000 is written. The program modules 1093 are stored, for example, in the hard disk drive 1090. For example, the program modules 1093 for executing processes similar to the functional configurations in the information provision systems 100 and 200 are stored in the hard disk drive 1090. The hard disk drive 1090 may be replaced by an SSD (Solid State Drive).

The setting data used in the processing of the above-described embodiment is stored as program data 1094, for example, in memory 1010 or hard disk drive 1090. Then, the CPU 1020 reads the program module 1093 or program data 1094 stored in memory 1010 or hard disk drive 1090 into RAM 1012 as necessary and executes it.

The program module 1093 and the program data 1094 may not necessarily be stored in the hard disk drive 1090, but may be stored in a removable storage medium, for example, and read by the CPU 1020 via the disk drive 1100 or the like. Alternatively, the program module 1093 and the program data 1094 may be stored in another computer connected via a network (such as a local area network (LAN) or wide area network (WAN)). The program module 1093 and the program data 1094 may then be read by the CPU 1020 from the other computer via the network interface 1070.

The above describes an embodiment of the invention made by the inventor, but the present invention is not limited to the description and drawings that form part of the disclosure of the present invention according to this embodiment. In other words, other embodiments, examples, operational techniques, etc. made by those skilled in the art based on this embodiment are all included in the scope of the present invention.

Reference Signs List 10, 210 Information providing device 11, 211 Prompt creation unit 12, 212 Summary generation unit 13 Expression conversion unit 20 Analyst terminal 30 Analysis server 40 Related party terminal 100, 200 Information providing system 121, 2121 Preprocessing model 122, 2122 Learning model 123, 2123 Evaluation model 131 Report material manager model 214 Extraction unit 2111 Prompt creation model 2141 Conversation manager model 2142 Dialogue summary model 2143 State detection model

Claims (6)

a prompt creation unit that creates a prompt for causing a language model to generate a sentence indicating an overview of a series of processes related to data analysis, based on execution parameters including parameters of each process and execution results of each process received from an analysis device that executes a data analysis tool that generates the series of processes related to data analysis;
an outline generation unit that generates a sentence indicating an outline of the series of processes by inputting the prompt generated by the prompt creation unit into the language model;
an expression conversion unit that converts a sentence indicating the summary generated by the summary generation unit into an expression suitable for a report and displays the converted sentence indicating the summary of the series of processes on a terminal device;
having
the language model is provided for each of the processes,
An information provision device characterized in that each language model is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the corresponding process and a sentence outlining the process created by an experienced analyst in response to the execution parameters of the corresponding process . The information providing device according to claim 1, characterized in that the expression conversion unit inputs each sentence indicating an overview of the process generated by each language model into a manager model, and generates a summary list in which each sentence is converted into an expression suitable for the report and associated with identification information of each process. an extraction unit that extracts, from a response to a dialogue between an analyst who has set a series of processes related to the data analysis and a dialogue model, an aim and/or ingenuity of the analyst with respect to the series of processes related to the data analysis;
and
The dialogue response includes a question of the dialogue model for eliciting the analyst's aim and/or ingenuity with respect to a series of processes related to the data analysis, and the analyst's answer in response to the question,
The information providing device according to claim 1, characterized in that the prompt creation unit creates a prompt for causing the language model to generate a sentence indicating an overview of the process by inputting the execution parameters received from the analysis device and the analyst's aim and/or ingenuity points extracted by the extraction unit into a prompt creation model. the summary generation unit inputs the prompt generated by the prompt creation unit into the language model provided for each of the processes;
The information providing device according to claim 3, characterized in that each language model has been trained in advance to generate a sentence outlining the process in accordance with the execution parameters of the corresponding process and the analyst's aims and/ or ingenuity. An information providing method executed by an information providing device, comprising:
A step of generating a prompt for causing a language model to generate a sentence indicating an overview of a series of processes related to data analysis, based on execution parameters including parameters of each process and execution results of each process received from an analysis device in which a data analysis tool that generates the series of processes related to data analysis is executed;
generating a sentence outlining the series of processes by inputting the prompt into the language model;
a step of converting the summary text generated in the generating step into an expression suitable for a report, and displaying the converted summary text of the series of processes on a terminal device;
Including,
the language model is provided for each of the processes,
An information provision method characterized in that each language model is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the corresponding process and a sentence outlining the process created by an experienced analyst in response to the execution parameters of the corresponding process . creating a prompt for causing a language model to generate a sentence indicating an overview of a series of processes related to data analysis, based on execution parameters including parameters of each process and execution results of each process received from an analysis device in which a data analysis tool that generates the series of processes related to data analysis is executed;
generating a sentence outlining the series of processes by inputting the prompt into the language model;
a step of converting the text indicating the summary generated in the generating step into an expression suitable for a report, and displaying the converted text indicating the summary of the series of processes on a terminal device;
on the computer ,
the language model is provided for each of the processes,
An information provision program in which each language model is a machine learning model in which machine learning is performed using as learning data a prompt based on the execution parameters of the corresponding process and a sentence outlining the process created by an experienced analyst in response to the execution parameters of the corresponding process .

Images