Organizational status assessment system, organizational status assessment device, organizational status assessment method and program

Abstract

This invention provides an organizational situation assessment system, an organizational situation assessment device, an organizational situation assessment method, and a program to implement these, which enable easy and effortless assessment of an organization's opinions and circumstances. [Solution] An organizational situation assessment system 1 for compiling individual opinions within an organization to create an organizational opinion and understand the organization's situation, comprising: an individual storage means 20 that extracts the words and actions and / or information of each individual from a specific recording means 100 and stores them in advance for each individual; a question means 30 that, when a predetermined question is created, asks the individual storage means 20 the question; an individual opinion extraction means 40 that, when a question is asked from the question means 30, extracts the wording that will be the answer to the question from the individual storage means 20 to create an individual opinion; and an aggregation means 50 that combines the individual opinions created by the individual opinion extraction means 40 to create an organizational opinion.

Description

[Technical Field]

[0001] The present invention relates to an organizational status assessment system, an organizational status assessment device, an organizational status assessment method, and a program for understanding opinions and other information within an organization. [Background technology]

[0002] Traditionally, in organizations such as companies, managers and other leadership personnel sometimes attempt to understand the organization's opinions and circumstances regarding certain questions or issues (see, for example, Reference 1). In the past, attempts were made to address this by conducting individual interviews and manually compiling the results to create a summary of the organization's opinions and circumstances. [Prior art documents] [Patent Documents]

[0003] [Patent Document 1] Japanese Patent Publication No. 2005-173948 [Overview of the project] [Problems that the invention aims to solve]

[0004] However, with the conventional methods described above, it is time-consuming and laborious to manually compile the results of each individual interview to create an organizational opinion and situation. As a result, it is practically difficult to compile the results of each individual interview, and consequently, it is difficult to grasp the organization's opinion and situation.

[0005] This invention has been made in view of the above problems, and aims to provide an organizational situation assessment system, an organizational situation assessment device, an organizational situation assessment method, and a program to implement these, which can easily and effortlessly grasp the opinions and circumstances of an organization. [Means for solving the problem]

[0006] To achieve the above objectives, the present invention provides an organizational situation assessment system for compiling the opinions of individuals within an organization to create an opinion for the organization as a whole and to grasp the situation of the organization, characterized in that the system comprises: personal storage means for extracting the words and actions and / or information of each individual from specific recording means and storing them in advance for each individual; questioning means for posing a predetermined question to each of the personal storage means when such a question is created; personal opinion extraction means for extracting the words that will be the answers to the question from each of the personal storage means when a question is posed from the questioning means, and creating an opinion for each individual; and aggregation means for compiling the opinions of each individual created by the personal opinion extraction means to create an opinion for the organization as a whole.

[0007] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal storage means is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records within the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal usage within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual, as part of an organizational situation awareness system.

[0008] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal opinion extraction means is an organizational situation assessment system that breaks down the question content to form subdivided decomposed question content, extracts wording that corresponds to the responses to the decomposed question content, and then uses the extracted wording to form an individual's opinion corresponding to the responses to the question content.

[0009] Furthermore, in addition to the configuration described above, the present invention is characterized in that the aggregation means is configured to aggregate the opinions of each individual, which have been created by the individual opinion extraction means, within a range classified by predetermined conditions, as part of an organizational situation assessment system.

[0010] Furthermore, in addition to the configuration described above, the present invention is characterized by an organizational situation understanding system configured to repeatedly generate organizational opinions using a large-scale language model, then perform deep learning using the records from the personal memory means, the extracted content from the personal opinion extraction means, and the aggregated results from the aggregation means in the repeated generation of organizational opinions, and then transition to generating organizational opinions from the deep learning.

[0011] Furthermore, the present invention relates to an organizational situation assessment device for controlling an organizational situation assessment system that aggregates the opinions of individuals within an organization to create an opinion for the organization and grasp the situation of the organization, characterized in that the organizational situation assessment device comprises: an individual storage means that extracts the words and actions and / or information of each individual from a specific recording means and stores it in advance for each individual; a questioning means that, when a predetermined question is created, poses the said question to each of the individual storage means; an individual opinion extraction means that, when a question is posed from the questioning means, extracts the wording that will be the answer to the said question from each of the individual storage means to create an opinion for each individual; and an aggregation means that aggregates the individual opinions created by the individual opinion extraction means to create an opinion for the organization.

[0012] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal storage means is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records within the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal usage within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

[0013] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal opinion extraction means is an organizational situation assessment device that breaks down the question content to form subdivided decomposed question content, extracts wording that corresponds to the responses to the decomposed question content, and then uses the extracted wording to form an individual's opinion corresponding to the responses to the question content.

[0014] Furthermore, in addition to the configuration described above, the present invention is characterized in that the aggregation means is an organizational status understanding device that aggregates the individual opinions created by the individual opinion extraction means within a range classified by specified conditions.

[0015] Furthermore, in addition to the configuration described above, the present invention is characterized by an organizational situation assessment device configured to repeatedly generate organizational opinions using a large-scale language model, then perform deep learning using the records from the personal memory means, the extracted content from the personal opinion extraction means, and the aggregated results from the aggregation means in the repeated generation of organizational opinions, and then transition to generating organizational opinions from the deep learning.

[0016] Furthermore, the present invention is a method for understanding the situation of an organization by aggregating the opinions of individuals within the organization to create an opinion for the organization, and is characterized by comprising: a personal storage step of extracting the words and actions and / or information of each individual from a specific recording means and storing them in advance in a personal storage means for each individual; a questioning step of posing a predetermined question to each of the personal storage means when the question is created; an individual opinion extraction step of extracting the words that will be the answers to the question from each of the personal storage means when the question is posed in the questioning step to create an opinion for each individual; and an aggregation step of aggregating the individual opinions created in the individual opinion extraction step to create an opinion for the organization.

[0017] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal storage step is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal usage within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

[0018] Furthermore, in addition to the configuration described above, the present invention is characterized in that the individual opinion extraction step is an organizational situation assessment method in which the question content is broken down to form subdivided decomposed question content, the words that will be responses to the decomposed question content are extracted, and then the extracted words are used to form an individual's opinion so that it corresponds to the response to the question content.

[0019] Furthermore, in addition to the configuration described above, the present invention is characterized in that the aggregation step is a method for understanding the organizational situation in which the opinions of each individual created in the individual opinion extraction step are aggregated within a range classified by specified conditions.

[0020] Furthermore, in addition to the configuration described above, the present invention is characterized by an organizational situation assessment method configured to repeatedly generate organizational opinions using a large-scale language model, then perform deep learning using the records of the personal memory means, the extracted content of the personal opinion extraction step, and the aggregated results of the aggregation step from the repeated generation of organizational opinions, and then transition to generating organizational opinions from the deep learning.

[0021] Furthermore, the present invention is a program that causes a computer to execute the processing of an organizational situation understanding system that collects the opinions of individuals in an organization to create an opinion for the organization as a whole and understand the situation of the organization, and is characterized in that the program causes a computer to execute the processing of the following: a personal storage step of extracting the words and actions and / or information of each individual from a specific recording means and storing them in advance in a personal storage means for each individual; a questioning step of posing a predetermined question to each of the personal storage means when the question is created; an individual opinion extraction step of extracting the words that will be the answers to the question from each of the personal storage means when the question is posed in the questioning step to create an opinion for each individual; and an aggregation step of collecting the individual opinions created in the individual opinion extraction step to create an opinion for the organization as a whole.

[0022] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal storage step is a program that causes a computer to perform a process of extracting and storing the words and / or actions and / or information of each individual from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal usage within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

[0023] Furthermore, in addition to the configuration described above, the present invention is characterized in that the personal opinion extraction step is a program that causes a computer to perform a process of breaking down the question content to form subdivided decomposed question content, extracting words that will be responses to the decomposed question content, and then forming an individual's opinion so that the extracted words correspond to the responses to the question content.

[0024] Furthermore, in addition to the configuration described above, the present invention is characterized in that the aggregation step is a program that causes a computer to perform aggregating the individual opinions created in the individual opinion extraction step within a range classified by predetermined conditions.

[0025] Furthermore, in addition to the configuration described above, the present invention is characterized by a program that causes a computer to perform deep learning using the records of the personal memory means, the extracted content of the personal opinion extraction step, and the aggregated results of the aggregation step, after repeatedly generating organizational opinions using a large-scale language model, and then transitioning from deep learning to generating organizational opinions. [Effects of the Invention]

[0026] According to the present invention, the system stores pre-extracted behaviors and information for each individual, and when a predetermined question is created, it extracts the appropriate responses from the stored behaviors and information to create individual opinions, which are then compiled to create an organizational opinion. This makes it easy to summarize the organization's opinion on a question. Furthermore, because individual opinions are created from pre-extracted behaviors and information, there is no need to conduct interviews or similar methods to answer the questions, allowing for the organization's opinion to be compiled conveniently and without hassle.

[0027] Furthermore, according to the present invention, in addition to records of individual interviews, it is possible to extract information about speech and actions in advance from records of meetings and terminals, records of transmissions on the network, image and video records, audio records, etc., so that more information about an individual's speech and actions can be obtained, and an individual's opinion in response to a question can be formulated more accurately.

[0028] Furthermore, according to the present invention, since the question content is broken down and the words corresponding to the broken-down question content are extracted, it is possible to extract individual opinions on the question in more detail and to create even more accurate individual opinions.

[0029] Furthermore, according to the present invention, aggregation can be performed under various conditions, and opinions from various organizations and situations can be generated with a single question.

[0030] Furthermore, according to the present invention, opinions generated using a large-scale language model can be stored, and these can be used to transition to generating opinions using deep learning, or opinions can be generated using deep learning simultaneously, thereby enabling the creation of more diverse and accurate opinions. [Brief explanation of the drawing]

[0031] [Figure 1] This is a functional block diagram illustrating the outline of an organizational status monitoring system according to an embodiment of the present invention. [Figure 2] This flowchart shows the main flow of the organizational status assessment system according to the said embodiment. [Figure 3] This is a functional block diagram showing the relationships between each phase of the organizational status assessment system according to the said embodiment. [Figure 4] This flowchart shows the control flow of the individual opinion extraction means of the organizational status assessment system according to the same embodiment. [Figure 5] This figure shows an example of organizational relationships in the organizational status assessment system according to the same embodiment. [Modes for carrying out the invention]

[0032] Embodiments of the present invention will be described below.

[0033] Figures 1 to 5 show embodiments of the present invention.

[0034] The organizational situation assessment system 1 of this embodiment is a system that gathers individual opinions within an organization such as a company to create an organizational opinion and grasp the organization's situation. Specifically, it collects in advance the words, actions, and information of individuals regarding their opinions in their daily work, etc., within the range of acquisition possible on a network N, etc., and stores them for each individual. For example, when a question is created by a designated person involved in the operation of the organization, such as a manager, director, or officer, the system extracts individual opinions on the question from the words, actions, and information collected in advance, without conducting direct responses such as interviews or questionnaires, aggregates the extracted individual opinions, and creates an organizational opinion on the question.

[0035] The scope from which information and actions can be obtained is expected to include, for example, terminals and servers on the company network, accessible servers on the internet, and interactions on social media. It is preferable to obtain information from many of these sources, but in some cases, it may be acceptable to obtain information only from appropriately restricted sources. For example, if the goal is to create a report based only on the opinions of a particular department within an organization, information could be obtained only from the network and equipment related to that department.

[0036] Furthermore, the term "individual" here may include, for example, full-time employees, contract employees, part-time workers, and temporary workers belonging to a company, as well as external individuals such as people from business partners. In addition, the target may be limited to individuals within an appropriate scope, or it may be broadly limited to the general public without specifying a scope; the scope should be determined as appropriate depending on the purpose and circumstances.

[0037] Furthermore, as shown in Figure 1, the organizational status monitoring system 1 of this embodiment consists of a server 10 as an organizational status monitoring device connected to a network N, a terminal 100 as a recording means, etc., and the server 10 as an organizational status monitoring device has a configuration that includes a personal storage means 20, a questioning means 30, a personal opinion extraction means 40, an aggregation means 50, etc.

[0038] Of these, the personal storage means 20 extracts either or both of each individual's words and actions and information from a specific recording means 100 and stores them in advance for each individual. Specifically, the personal storage means 20 is configured to extract and store each individual's words and actions and information from any or all of the following as records within the specific recording means 100: records of interviews with the individual, records of meetings attended by the individual, records of terminal usage within the organization, records of outgoing calls on the network, records of images and videos involving the individual, and records of audio involving the individual. Here, the personal storage means 20 is set up for each individual, and the data of that individual is stored in the respective personal storage means 20. It is also possible that the personal storage means 20 is configured to have a storage area for each individual within its framework.

[0039] Furthermore, information is acquired as needed from terminals 100 connected to network N. While text information is preferred for the collected data, image information, video information, audio information, or any other appropriate information, or a combination of these, is also acceptable. For example, the system should extract information that appears to relate to a specific person from email bodies and attachments, social media conversations, internal company documents on the network such as servers, and word-of-mouth. This information extraction may be filtered using AI. The personal memory device 20 is configured to collect this personal information simultaneously from multiple sources depending on the time and circumstances.

[0040] Furthermore, regarding the methods of collecting information on each individual's words and actions, it is acceptable to collect as much information as possible without specifying any particular field or type, or to pre-determine the fields and types to be collected and collect words and actions that conform to those categories. In addition, when words and actions and information are collected, all the information may be stored in a single memory area, or categories may be established when storing the information, and memory areas may be divided according to the field of information.

[0041] Furthermore, the questioning means 30, when a manager, director, or officer who wishes to know a certain situation within the organization has created a predetermined question, sends that question to the respective personal memory means 20. Here, the question itself may be created within the server 10, or it may be created on an external terminal and then moved to the questioning means 30 within the server 10. Also, the question may be created by text input, voice input, or by converting input such as image recognition into text.

[0042] Furthermore, when a question is posed by the questioning means 30, the individual opinion extraction means 40 extracts the appropriate responses from each individual memory means 20 to create individual opinions. Here, the individual opinion extraction means 40 is configured to break down the content of the question from the questioning means 30 to form subdivided question content, extract the appropriate responses to these subdivided question content, and then use the extracted text to form individual opinions corresponding to the responses to the questions. The processing performed by the individual opinion extraction means 40 is categorized as the "information extraction phase."

[0043] Furthermore, the aggregation means 50 combines the individual opinions created by the individual opinion extraction means 40 to create an organizational opinion. Here, the aggregation means 50 aggregates the individual opinions created by the individual opinion extraction means 40 within a range classified by specified conditions. For example, when creating an opinion for the entire company, it is configured to aggregate based on the opinions of all individuals belonging to that department. When creating an opinion for department α (or β) as shown in Figure 5, it aggregates only the opinions of individuals belonging to that department α (or β). Similarly, when creating an opinion for a subordinate department α-1 or α-2 (or β-1 or β-2), it aggregates only the opinions of individuals belonging to that department. In addition, when aggregating only the opinions of individuals at the manager level or higher, it aggregates only the opinions of individuals a to g, and so on, allowing the aggregation work to be performed by setting an appropriate aggregation range. It is also possible to perform aggregation for multiple ranges simultaneously and display the results in parallel. The processing performed by the aggregation means 50 is categorized as "Information Aggregation Phase".

[0044] In this embodiment, after extracting individual opinions without weighting, weights may be applied according to each individual's information. Here, the weighting may be based on the reliability of the information source. For example, the reliability of the information may change depending on whether the individual is a highly reliable or unreliable person, and weighting may be applied according to the information source. In addition, even if similar opinions are obtained, weighting may be applied because the credibility may differ depending on the information source. After applying the weights as described above, the importance of each individual's opinion may be set (for example, assigning different weights to the same opinion for each individual). The weighting may vary depending on the person, organization, external service, etc. of the information source. Furthermore, the weighting may also take into account the time axis, region, location, etc.

[0045] In this embodiment, a large-scale language model (LLM) was used to extract individual opinions and then aggregate those opinions. However, the system is not limited to this, and other methods may be used for extracting and aggregating opinions. Furthermore, the system may be configured to improve learning accuracy by repeatedly extracting individual opinions and creating organizational opinions using the large-scale language model, and then perform deep learning using the records from the personal memory means 20, the extracted content from the individual opinion extraction means 40, and the aggregation results from the aggregation means 50 from the repeated creation of organizational opinions, before transitioning to creating organizational opinions from the deep learning.

[0046] Furthermore, in this embodiment, these results are summarized and configured to be displayed on the screen in a way that visualizes the organization's opinions using tables, graphs, etc.

[0047] Next, the main flow of the organizational status assessment method in the organizational status assessment system 1 (organizational status assessment device 10) of this embodiment will be explained using Figure 2.

[0048] Here, employee m as shown in Figure 5 a ~m s Using the company where [Name] works as a model, we will specifically explain the process (organizational situation assessment method) of generating and aggregating answers to question Q posed by President A using organizational situation assessment system 1, using Figures 2-4.

[0049] First, collect information (for example, collect information in text) through so-called 1on1 interviews (interviews conducted one-on-one between a supervisor and a subordinate), etc., and store this information in the personal memory means (accumulated text data) 20. This information may be collected multiple times regularly. In that case, all the information may be accumulated and stored, or when the opinions on the same content have changed (for example, when the opinion that there was dissatisfaction with welfare benefits before has changed to the opinion that one is satisfied with welfare benefits recently, etc.), it may be overwritten with the latest one. In this embodiment, such a personal memory means 20 is used, and later, an answer to a question by the questioning means 30 is obtained by the personal opinion extraction means 40, and information is extracted and processed accordingly. In this way, the advantage of being able to set highly flexible questions is obtained. And these personal opinions are summarized by the aggregation means 50.

[0050] Details of the collection and storage of speech and information in the personal memory means 20 will be described.

[0051] Here, for employee m a ~m s if there is supplementary information c a ~c s exists, it is stored in the data taking this supplementary information into account. For example, check whether there is such text data as belonging, position, the personality and human relations of the person himself (if he is a intimidating person (in this case, it is supplemented that the speech of the person sitting next to him is likely to be speculative), hates Mr. m x (in this case, it is supplemented that the bad evaluation of Mr. m x should be regarded as nonsense)), etc., and if it exists, record to that effect as a supplement.

[0052] Next, check whether the interview records T1 to T n for which the speaker has been identified already exist. At the time when President a asks (creates) question Q, there are n interview records T1, T2,,, T n in the form in which the speaker has been identified and accumulated in the company (organization). Here, the y-th interview record Ty The list of participants in P(T) y ) is represented as P(T y ) a certain employee m x T y (m x It will be represented as ). For example, the m that was carried out in α-1 i and m j Considering the fifth interview record T5 in which [the person] participated, P(T5) = [m i ,m j ] and within that, m j The statement was made by T5(m i ) and similarly m j The statement was made by T5(m j This is expressed as )

[0053] First, as shown in Figure 2, the user (in this case, President a) inputs a question Q about the organizational situation they want to understand, for example, using the questioning means 30 (step S1). At this point, the user can also specify conditions for the text to be searched (for example, the search period). The search conditions specified here are denoted as F.

[0054] Next, the individual opinion extraction means 40 decomposes and classifies the questions of the question means 30 (step S2). Here, question Q is taken as input, and prompts that fit the template shown below are input to the LLM, thereby further decomposing question Q into q1, q2,,,q z To obtain.

[0055] As a prompt template, in order to break down the question "Q" about the organization into smaller questions, 1. The implicit assumptions that underlie the question in question, 2. Questions explicitly included in the question, We extracted information that falls under these two perspectives, organized it in a MECE (Mutually Exclusive, Collectively Exhaustive) manner, and addressed each question. • Determining whether the question is a closed question that can be answered with YES / NO or an open question that cannot. • Calculate the degree of confidence that the question is indeed contained within the original question before it was divided. After performing the above steps, the program is instructed to return the output in JSON format. It is also instructed not to output any format other than JSON.

[0056] The output will be q1, q2, ..., q, obtained by splitting the original question Q into MECE (Mutually Exclusive, Collectively Exhaustive) parts. z The output is as follows: Each 'q' is structured to include the question, closed / open classification, confidence level regarding inclusion, etc.

[0057] Next, the question prompt is generated (step S3).

[0058] The output q1, q2, , q in step S2 above z Based on this, the corresponding question prompts p1, p2,,,p z This generates the decomposed questions q1, q2,,,q z Each closed / open classification (type) is, • If it's a closed question → Closed question prompt • If it's an open question → Open question prompt They are designed to be used in their respective forms.

[0059] Note that in the following sections, q1, q2,,,q z The i-th question q i An example of a prompt is shown. The following are used as placeholders (markers for embedding corresponding text later). target_person: Embed information indicating who you want to extract opinions from. existing_tags: q tags extracted from the prior analysis i A unique list of answer tags. Embed it.

[0060] For a closed-question prompt, generate something like this: target_person's q i (The following questions) are to be extracted and output in a format conforming to the JSON format described below. • Voices: Statements that suggest what opinion the person holds regarding the question. • Summary: A qualitative summary of the target person's stance on the questions, based on the voices. • Stance: How strongly you feel "YES" or "agree" (indicated on a scale of 0 to 1, where 1 means "completely YES" or "completely agree," and 0 means "completely NO" or "absolutely reject"). • Confidence: Reliability of the analysis results (expressed as a value between 0 and 1, based on the idea that 1 minus the probability that the analysis results are incorrect).

[0061] Note that the q of target_person i Since the target text may not contain information that suggests an opinion on the subject, for items where information extraction from the text is difficult, the system should not force the generation of the corresponding item, but instead instruct it to return an empty value (or 0) that is appropriate for the format of that item.

[0062] The following is generated as an open-question prompt: If target_person is q from the target text i To extract the likely opinions regarding (the following questions), the analysis will be processed step-by-step in the order shown below. 1. Extraction of voices (statements that suggest what opinions people have on the question) 2. Generation of summary (a summary based on the voices of the target person found in the text) 3. Evaluation of summary confidence (the probability that the summary is appropriate based on the text) 4. Extraction of tags (the 5W1H information elements that constitute the answers to the questions extracted from the summary) 5. Evaluation of the confidence of each tag (the probability that it is appropriate to tag the target person's opinion).

[0063] Please note the following points when outputting the analysis results. When extracting items, use the existing expressions / notations within existing_tags for any expressions that can be replaced by items within existing_tags. • Output should be in JSON format as described below; no output other than JSON should be provided. • Since the text may not contain opinions on the target_person's questions, for items where information extraction from the text is difficult, do not force the generation of the corresponding item, but instead return an empty value (or 0) that is appropriate for the format of that item.

[0064] Next, we identify the text to be searched (step S4). T1~T n Identify those that meet the search condition F. For simplicity of explanation, here we will refer to T1~T n All of the above conditions must be met.

[0065] Next, identify the texts to be tested for each individual (Step S5). Here, each interview record T1-T n Participant information P(T1)~P(T n ) is obtained. Based on that information, each m a~s Interview records t(m a )~t(m s ) to identify a certain person m. x The y-th interview record among the interview records involved is t(m x It is written as ,y).

[0066] Next, we extract the responses (=information) from each individual (Step S6). Here, as shown in Figure 4, first, we extract the next person (=m x The process of extracting information from ) is started (step S21).

[0067] Next, m xInterpretation and considerations regarding this are specified (Step S22). Here, the person m as described above x Supplementary information C x If present, the following prompt will specify the interpretation notes for the text. Note that the items listed here are just examples, and are included in Supplementary Information C. x Supplemental information C consisting of information other than the following, or containing information of other content. x That's fine too. prompt Now, let's talk about "Q" x Analyze the text which may contain the opinion of m x Extract the opinions. In that case, the following m x Characteristics (= Supplementary Information C) x Interpretations should be made based on the content of [the relevant section]. They tend to jump to conclusions and make definitive statements based on assumptions. I have a very bad relationship with [Name of person]. • It has the characteristics of an opposition party. They tend to pretend to know everything. He is a yes-man who follows the orders of his superiors. They are the type of person who finds it difficult to express negative opinions, perhaps because they don't want to hurt others' feelings. They don't care about anything that doesn't directly affect their own interests. • Having been dispatched from the United States, they are confused and unfamiliar with Japanese culture.

[0068] Next, m x Interview records involved t(m x ) obtain (step S23).

[0069] Next, the following interview record t(m x Specify p1, p2,,,p obtained in step S3 as the target of analysis (step S24). z Fill in the placeholder with the value you should actually use. ·target_person:m x • existing_tags: A unique list of tags extracted for each corresponding question.

[0070] Next, extract and record the answers to each question (step S25). The placeholders p1, p2,,,p are filled with values. z Using m in the text to be analyzed x We will extract opinion information from q1, q2,,,q z The kth question q within a certain divided question. k t(m x Opinion information extracted from ,y) a k (m x Let it be expressed as ,y). Here, a k (m x If ,y) is used, update existing_tags based on the response result.

[0071] Next, t(m x Check whether the process has been completed to the end (step S26). If there are still unprocessed interview records, return to step S24.

[0072] If the process was completed in step S26, the next step is to integrate and record the individual opinion information (step S27). q1, q2,,,q z For each of the divided questions, the following prompt is used to ask the individual m x Integrate and record the opinions and information. Also, the kth question q k m for x Integrated opinion information A k (m x )

[0073] Prompt: kth question q k Integration prompt if it was a closed question "q k Regarding the question, "m xOutput opinion information. Output other than the specified JSON format is unnecessary. Note that the opinion information output will be based solely on the input JSON listed below, and no other information or speculation will be used. • Voices: Statements that suggest what opinion the person holds regarding the question. ·summary:Based on voices x A qualitative summary of the stance on the question. • Stance: How strongly you feel "YES" or "agree" (indicated on a scale of 0 to 1, where 1 means "completely YES" or "completely agree," and 0 means "completely NO" or "absolutely reject"). • Confidence: Reliability of the analysis results (expressed as a value between 0 and 1, based on the idea that 1 minus the probability that the analysis results are incorrect).

[0074] Prompt: kth question q k Integration prompt if it was an open question "q k Regarding the question, "What is the opinion?", the following opinion information will be output in the JSON format shown below. Output other than the specified JSON format is not required. Note that the opinion information will be based solely on the input JSON listed below, and no other information or speculation will be used. In addition, the output tags will only use tags that appear in the input JSON, and no new tags will be generated. 1. Extraction of voices (statements that suggest what opinions people have on the question) 2. summary(m found in the text) x Generation of a summary based on the voices of opinion 3. Evaluation of summary confidence (the probability that the summary is appropriate based on the text) 4. Extraction of tags (the 5W1H information elements that constitute the answers to the questions extracted from the summary) 5. confidence(m x Evaluation of the probability that the tag is appropriate for the opinion.

[0075] Note that at this stage, input filtering based on confidence may be performed by giving a prompt such as "Do not use information with a confidence of less than 0.7".

[0076] Next, it is checked whether the processing has reached the last person or the same task (step S28). Here, if there are still unprocessed persons, the process returns to step S21.

[0077] If the processing has reached the last person in step S28, the extraction completion notification shown in FIG. 2 is given (step S7), and then, the input of aggregation conditions is performed (step S8). As the aggregation conditions, for example, · Belonging: Section α-2, etc. · Rank: Section chief or higher, etc. · Response status to specific questions: Answered, with a YES degree of 0.5 or more for question θ, etc. Conditions such as these are input. Let the conditions specified here be G.

[0078] Next, the individuals to be aggregated are specified (step S9). Here, the persons corresponding to G are specified. For the sake of simplicity of explanation, it is assumed that m h ~m s are applicable.

[0079] Next, the responses of the persons to be aggregated are searched (step S10). When the question Q is decomposed into q1, q2,,, q z assuming that A1(m h~s )~A z (m h~s ) are searched and listed.

[0080] Next, charting and summary generation are performed (step S11). A1(m h~s )~A z (m h~sFor closed questions, it holds stance (0 - 1), and for open questions, it holds tags as items that can be aggregated respectively. Therefore, graphing and table generation can be performed using this. Also, since each has a qualitative summary part called summary, a summary text can be generated after listing these. Let the summarized answer for the question Q generated here be S Q be.

[0081] Note that at this stage, input filtering by confidence may be performed by giving a prompt such as "Do not use information with confidence less than 0.7".

[0082] Next, the transition to the Deep Learning (DL) model will be explained. First, regarding the transition to the DL model for extracting individual question - answers from text, in "extracting each individual's answer (= information)", it is also possible to construct a more compact and specialized Deep Learning model rather than the LLM by using, as input data and teacher data respectively, the following elements output by the LLM for learning, and one can use that. <00,00435>As input data, individual interview records t(m x , y), prompt p k As teacher data, opinion information a k corresponding to the prompt p extracted from the interview record k (m x , y)

[0083] Also, regarding the transition to the DL model for generating integrated individual opinions from individual question - answers, it is also possible to construct a more compact and specialized Deep Learning model rather than the LLM by using, as input data and teacher data respectively, the following elements output by the LLM in step S27 for learning, and one can use that. As input data, opinion information a k corresponding to the prompt p extracted from the interview record k (m x,1~) As training data, question q k Individual integrated opinion A k (m x )

[0084] Furthermore, regarding the transition to a DL model for generating the final summary from individual opinions, it is also possible to construct a more compact and specialized Deep Learning model instead of LLM by training the following elements output by LLM in step S11 as input data and training data, respectively, and this can also be used. The input data consists of individual integrated opinions A1~(m1~) regarding question q1~. The training data consists of summarized answers S to question Q. Q

[0085] As described above, according to this embodiment, the actions and information extracted for each individual are stored in advance, and when a predetermined question is created, the words that will be the answers to that question are extracted from the stored actions and information to create individual opinions, and these are then combined to create an organizational opinion. Therefore, the organization's opinion on a question can be easily compiled. Furthermore, since individual opinions are created from actions and information stored in advance, there is no need to conduct interviews or other methods to answer the questions, and the organization's opinion can be compiled in a convenient and hassle-free manner.

[0086] Furthermore, according to this embodiment, in addition to records of individual interviews, it is possible to extract information about speech and actions in advance from records of meetings and terminals, records of calls made on the network, image and video records, audio records, etc., so that more information about an individual's speech and actions can be obtained, and an individual's opinion in response to a question can be formulated more accurately.

[0087] Furthermore, according to this embodiment, since the question content is broken down and the wording corresponding to the broken-down question content is extracted, it is possible to extract individual opinions on the question in more detail and to create even more accurate individual opinions.

[0088] Furthermore, according to this embodiment, aggregation can be performed under various conditions, and opinions from various organizations and situations can be generated with a single question.

[0089] Furthermore, according to this embodiment, opinions created using a large-scale language model can be stored, and these can be used to transition to generating opinions using deep learning, or deep learning can be used to generate opinions simultaneously, thereby enabling the creation of more diverse and accurate opinions.

[0090] Furthermore, the present invention is not limited to the embodiments described above, but can be applied to other configurations and other usage situations.

[0091] For example, the above-described embodiment described an organizational situation monitoring system 1 that operates as a system connected via the Internet, but it is not limited to this, and it may also be an organizational situation monitoring device that performs everything from collecting individual behaviors and information to extracting individual opinions, aggregating them, and outputting the results, all in a single device. It may also be a program for realizing the contents of the present invention, and in this case, the program may be applied to an appropriate system or device to realize the invention. Furthermore, the system is not limited to one that operates via the Internet, but may also be established via other networks (intranet, closed network, etc.). Furthermore, the network may be a system connected by wire and / or wireless. [Explanation of Symbols]

[0092] 1. Organizational Status Assessment System 10. Organizational status monitoring device 20 Personal Storage Means 30 Question methods 40 Personal opinion extraction means 50 Aggregation methods 100 Recording means N Network

Claims

1. An organizational situation assessment system that collects individual opinions within an organization to create an opinion for the organization as a whole and grasps the situation of the organization, A personal storage means that extracts the words and / or information of each individual from a specific recording means and stores it in advance for each individual, When a predetermined question is created, a questioning means poses the said question to each of the aforementioned personal memory means, When a question is posed by the aforementioned questioning means, the personal opinion extraction means extracts the wording that will be the response to the question from each of the aforementioned personal memory means and creates an opinion for each individual, A means for compiling the individual opinions created by the individual opinion extraction means to create an organizational opinion, An organizational status assessment system characterized by having the following features.

2. The organizational situation monitoring system according to claim 1, characterized in that the personal storage means is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal use within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

3. The organizational situation assessment system according to claim 1, characterized in that the personal opinion extraction means breaks down the question content to form subdivided sub-question content, extracts wording that corresponds to the responses to the sub-question content, and then forms an individual opinion using the extracted wording to correspond to the responses to the question content.

4. The organizational status assessment system according to claim 1, characterized in that the aggregation means performs aggregation on the individual opinions created by the individual opinion extraction means within a range classified by specified conditions.

5. The organizational situation understanding system according to claim 1, characterized in that, after repeatedly generating organizational opinions using a large-scale language model, deep learning is performed using the records of the personal memory means, the extracted content of the personal opinion extraction means, and the aggregated results of the aggregation means in the repeated generation of organizational opinions, and then the system transitions to generating organizational opinions from the deep learning.

6. An organizational status assessment device that controls an organizational status assessment system that collects individual opinions within an organization to create an opinion for the organization and grasps the status of the organization, A personal storage means for extracting the words and actions and / or information of each of the aforementioned individuals from a specific recording means and storing them in advance for each of the aforementioned individuals, When a predetermined question is created, a questioning means poses the said question to each of the aforementioned personal memory means, When a question is posed by the aforementioned questioning means, the personal opinion extraction means extracts the wording that will be the response to the question from each of the aforementioned personal memory means and creates an opinion for each individual, A means for compiling the individual opinions created by the individual opinion extraction means to create an organizational opinion, An organizational status assessment device characterized by having the following features.

7. The organizational status monitoring device according to claim 6, characterized in that the personal storage means is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal use within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

8. The organizational situation assessment device according to claim 6, characterized in that the personal opinion extraction means breaks down the question content to form subdivided sub-question content, extracts wording that corresponds to the responses to the sub-question content, and then forms an individual opinion using the extracted wording to correspond to the responses to the question content.

9. The organizational status assessment device according to claim 6, characterized in that the aggregation means performs aggregation on the individual opinions created by the individual opinion extraction means within a range classified by specified conditions.

10. The organizational situation understanding device according to claim 6, characterized in that, after repeatedly generating organizational opinions using a large-scale language model, deep learning is performed using the records of the personal memory means, the extracted content of the personal opinion extraction means, and the aggregated results of the aggregation means in the repeated generation of organizational opinions, and then the device transitions to generating organizational opinions from the deep learning.

11. A method for understanding the situation of an organization, which involves compiling the opinions of individuals within an organization to create an opinion representing the organization as a whole, and understanding the situation of the organization as a whole. A personal storage step involves extracting the words and / or information of each individual from a specific recording means and storing it in a personal storage means for each individual in advance. When a predetermined question is created, the question is posed to each of the aforementioned personal memory devices; When a question is asked in the questioning step, the personal opinion extraction step extracts the wording that will be the answer to the question from each of the personal memory means and creates an opinion for each individual, The process involves aggregating the individual opinions created in the individual opinion extraction process to create an organizational opinion, A method for understanding the organizational status, characterized by having the following features.

12. The method for understanding the organizational situation according to claim 11, characterized in that the personal memory step is configured to extract and store the words and / or actions and / or information of each individual from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal use within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

13. The method for understanding the organizational situation according to claim 11, characterized in that the individual opinion extraction step involves breaking down the question content to form subdivided sub-questions, extracting phrases that are responses to the sub-questions, and then forming individual opinions using the extracted phrases so that they correspond to the responses to the questions.

14. The method for understanding the organizational situation according to claim 11, characterized in that the aggregation step is performed on the individual opinions created in the individual opinion extraction step, within a range classified by specified conditions.

15. The method for understanding the organizational situation according to claim 11, characterized in that, after repeatedly generating organizational opinions using a large-scale language model, deep learning is performed using the records of the personal memory means, the extracted content of the personal opinion extraction process, and the aggregated results of the aggregation process in the repeated generation of organizational opinions, and then the method transitions to generating organizational opinions from the deep learning.

16. A program that causes a computer to perform the processing of an organizational situation assessment system, which collects the opinions of individuals within an organization, creates an opinion for the organization as a whole, and grasps the situation of the organization, A personal storage step involves extracting the words and / or information of each individual from a specific recording means and storing it in a personal storage means for each individual in advance. When a predetermined question is created, the question is posed to each of the aforementioned personal memory devices; When a question is asked in the questioning step, the personal opinion extraction step extracts the wording that will be the answer to the question from each of the personal memory means and creates an opinion for each individual, The process involves aggregating the individual opinions created in the individual opinion extraction process to create an organizational opinion, A program characterized by causing a computer to perform a process that has the following characteristics.

17. The program according to claim 16, wherein the personal memory step causes a computer to perform a process of extracting and storing information about each individual's words and actions from any or all of the following as records in the specific recording means: records of interviews with the individual, records of meetings attended by the individual, records of terminal use within the organization, records of outgoing calls on the network, images and / or video records involving the individual, and audio records involving the individual.

18. The program according to claim 16, wherein the personal opinion extraction step involves breaking down the question content to form subdivided sub-question content, extracting wording that corresponds to the answers to the sub-question content, and then causing a computer to perform a process of forming an individual opinion using the extracted wording so that it corresponds to the answers to the question content.

19. The program according to claim 16, characterized in that the aggregation step causes a computer to perform a process of aggregating the individual opinions created in the individual opinion extraction step within a range classified by specified conditions.

20. The program according to claim 16, characterized in that, after repeatedly generating organizational opinions using a large-scale language model, the program causes a computer to perform deep learning using the records in the personal memory means, the extracted content of the personal opinion extraction process, and the aggregated results of the aggregation process in the repeated generation of organizational opinions, and then transitions to generating organizational opinions from the deep learning.

Images