Consensus-building support system and consensus-building support method

Abstract

To provide a consensus-building support system and method that analyzes the context of opinions from stakeholders and automatically addresses emotional opposition and distrust. [Solution] The system is characterized by comprising: an information processing device; an input means for inputting the opinions of relevant parties to the information processing device; an analysis unit that analyzes the opinions input by the input means using an emotion recognition engine and calculates the emotional state as an emotion score; an estimation unit that estimates the consensus formation stage from the emotion score analyzed by the analysis unit; a dynamic dialogue generation unit that automatically generates dialogue content based on the consensus formation stage estimated by the estimation unit for the opinions and transmits the dialogue content to the person who expressed the opinion; a storage device that stores the history of the emotional state for each person who expressed the opinion, the consensus formation stage, and the dialogue content automatically generated by the dynamic dialogue generation unit; and a management unit that manages the history of the emotional state for each person who expressed the opinion, the consensus formation stage, and the dialogue content automatically generated by the dynamic dialogue generation unit stored in the storage device.

Description

Technical Field

[0001] The present invention mainly relates to a consensus formation support system and a consensus formation support method for assisting in smoothly reaching an agreement with interested parties such as residents in the construction of so-called unpopular facilities such as waste treatment plants, crematoria, nuclear power facilities, and soil pollution countermeasures.

Background Art

[0002] Conventionally, as a system for exchanging opinions with residents in the construction of unpopular facilities, etc., there is a development project operation system that connects a resident-side terminal that can be used by local residents in a development project such as the construction of a treatment plant or disposal site, and a server that shares and manages opinions such as the desires and claims of residents via a communication network. The server includes an opinion information registration means for registering the opinion information of residents transmitted from the resident-side terminal in an opinion information database, and an opinion information management means for processing the opinion information of residents registered by this opinion information registration means and managing it so that it can be efficiently utilized at each business stage.

[0003] However, in such a development project operation system, although it is possible to register opinions from interested parties such as residents, it is impossible to understand the context of those opinions, and it is impossible to automatically respond to emotional opposition or distrust. There was a problem that it had to rely on experts for responses, etc.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] In view of the aforementioned shortcomings of conventional systems, the present invention aims to provide a consensus-building support system and method that analyzes the context of opinions from stakeholders and automatically responds to emotional opposition and distrust. [Means for solving the problem]

[0006] To achieve the above objective, the consensus-building support system described in claim 1 of the present invention includes: an information processing device; an input means for inputting the opinions of stakeholders into the information processing device; an analysis unit that analyzes the opinions input by the input means using an emotion recognition engine and calculates the emotional state as an emotion score; an estimation unit that estimates the consensus-building stage from the emotion score analyzed by the analysis unit; a dynamic dialogue generation unit that automatically generates dialogue content based on the consensus-building stage estimated by the estimation unit for the opinions and transmits the dialogue content to the person who expressed the opinion; a storage device that stores the emotional state of each person who expressed the opinion, the consensus-building stage, and a history of the dialogue content automatically generated by the dynamic dialogue generation unit; and the storage device. The system comprises an analysis unit, an estimation unit, a dynamic dialogue generation unit, a storage device, and a management unit that manages the history of the emotional state of each person who expressed the opinion stored in the system, the consensus-building stage, and the dialogue content automatically generated by the dynamic dialogue generation unit. The analysis unit, estimation unit, dynamic dialogue generation unit, storage device, and management unit are provided in the information processing device or connected to the information processing device via an internet connection. The estimation unit calculates a negative emotional score by weighting the emotional scores for anger, distrust, and anxiety with predetermined weights based on the emotional score calculated by the analysis unit, smooths the calculated negative emotional score over time by applying an exponential moving average, and estimates the consensus-building stage based on predetermined thresholds and consecutive occurrence conditions.

[0007] The emotion recognition engine of the analysis unit of the consensus building support system according to claim 2 uses natural language processing and a large-scale language model to analyze the context of the opinion and calculate a plurality of emotion categories as multidimensional vectors, and the dynamic dialogue generation unit automatically adjusts the dialogue content to be generated according to the direction and magnitude of the multidimensional vectors calculated by the emotion recognition engine.

[0008] The management unit of the consensus-building support system according to claim 3 includes an attribute estimation module that analyzes the speaker's speech history and behavioral patterns and estimates attributes, provides the attributes analyzed by the attribute estimation module to the dynamic dialogue generation unit, and the dynamic dialogue generation unit further refers to the attributes to automatically generate dialogue content.

[0009]

[0010] The attribute estimation module of the consensus-building support system described in claim 4 analyzes the relationship between the sentiment score changes, keyword frequency, affiliated organizations, family structure, educational background, and relevant news reporting period of the person who input the opinion, and classifies the area of ​​interest, and the dynamic dialogue generation unit adjusts the information presented and the style of expression of the dialogue content based on the estimated area of ​​interest and social background.

[0011] The consensus-building support method described in claim 5 comprises: an opinion acquisition step of acquiring opinions from stakeholders; an emotion analysis step of analyzing the opinions acquired in the opinion acquisition step using an emotion recognition engine and calculating the emotional state as an emotion score; a consensus-building stage estimation step of estimating the consensus-building stage from the emotion score analyzed in the emotion analysis step; a dynamic dialogue generation step of automatically generating dialogue content and transmitting the dialogue content to the person who expressed the opinion, based on the consensus-building stage estimated in the consensus-building stage estimation step for the opinions acquired in the opinion acquisition step; and a management step of storing and centrally managing the emotional state of each person who expressed the opinion, the consensus-building stage, and the dialogue history automatically generated in the dynamic dialogue generation step in a storage device. The consensus-building stage estimation step is characterized in that, based on the emotion score, a negative emotion score is calculated by weighting the emotion scores for anger, distrust, and anxiety with predetermined weights, and the calculated negative emotion score is smoothed over time by applying an exponential moving average, and the consensus-building stage is estimated based on predetermined thresholds and consecutive occurrence conditions.

[0012] The emotion analysis step of the consensus building support method according to claim 6 is characterized in that the emotion recognition engine uses natural language processing and a large-scale language model to analyze the context of the opinion and calculate a plurality of emotion categories as multidimensional vectors, and the dynamic dialogue generation step is characterized in that the dialogue content to be generated is automatically adjusted according to the direction and magnitude of the multidimensional vectors calculated by the emotion recognition engine.

[0013] In the opinion acquisition step of the consensus-building support method described in claim 7, the attributes of the person who input the opinion are estimated using an attribute estimation module, and the dialogue content is automatically generated by further referencing the attributes, and the attribute estimation module analyzes the relationship between the sentiment score changes, keyword frequency, affiliated organizations, family structure, educational background, and relevant news reporting period of the person who input the opinion to classify areas of interest, and the dynamic dialogue generation Step This system is characterized by adjusting the information presented and the style of expression in the dialogue based on the estimated areas of interest and social background. [Effects of the Invention]

[0014] As is clear from the above explanation, the present invention provides the following effects. (1) In each of the inventions described in claims 1 to 9, the emotions of the person who entered the opinion can be quantified and classified by analyzing the opinion entered by the input means 3 using the emotion recognition engine 4 and scoring the emotional state. (2) Furthermore, by quantifying and classifying emotions, the progress of consensus building can be quantified and visualized. (3) It can estimate the consensus-building stage and automatically generate dynamic dialogue templates. (4) Because these responses can be performed automatically, it can reduce the workload on personnel and provide a standardized consensus-building process. (5) While conventional AI dialogue systems were limited to understanding the context of speech, by incorporating consensus-building algorithms (weighted average, EMA, threshold judgment) and linking them with attribute estimation, the progress of consensus-building can be quantified and visualized, enabling consensus-building support with transparency and accountability. (6) In the prior art, opinions were only registered and managed in a database. However, in the present invention, emotions are quantified as emotion scores, the consensus formation stage can be estimated, and dialogues can be automatically generated. Also, the dialogue content can be automatically generated based on attributes as needed. (7) Since the history of the consensus formation stage and the dialogue content automatically generated by the dynamic dialogue generation unit can be centrally managed, instead of only evaluating instantaneous utterances, time-series processing is introduced to enable stable estimation.

Brief Description of the Drawings

[0015] FIGS. 1 to 4 are explanatory diagrams showing a first embodiment of the present invention. [Figure 1] Block diagram of the consensus formation support system according to the first embodiment. [Figure 2] Process diagram of the consensus formation support method according to the first embodiment. [Figure 3] Flowchart of the emotion analysis step. [Figure 4] Flowchart of the consensus formation stage estimation step.

Modes for Carrying Out the Invention

[0016] Hereinafter, the present invention will be described in detail according to the modes for carrying out the present invention shown in the drawings.

[0017] In the first mode for carrying out the present invention shown in FIGS. 1 to 4, 1 is a consensus formation support system mainly for smoothing the consensus formation among stakeholders such as residents in so-called NIMBY-type cases (Not In My Backyard) such as waste treatment plants, crematoria, nuclear power facilities, and soil pollution countermeasures.

[0018] As shown in FIG. 1, this consensus formation support system 1 includes an information processing device 2, an input means 3 capable of inputting opinions of related parties such as residents to the information processing device 2, an analysis unit 5 that analyzes the opinions input by the input means 3 using an emotion recognition engine 4 and calculates the emotional state as an emotion score, an estimation unit 6 that estimates the consensus formation stage from the emotion score analyzed by the analysis unit 5, a dynamic dialogue generation unit 7 that automatically generates dialogue content based on the consensus formation stage estimated by the estimation unit 6 for the opinions and transmits the dialogue content to the person who issued the opinions, a storage device 8 that stores the history of the emotional state, the consensus formation stage, and the dialogue content automatically generated by the dynamic dialogue generation unit for each person who issued the opinions, and a management unit 9 that manages the history of the emotional state, the consensus formation stage, and the dialogue content automatically generated by the dynamic dialogue generation unit for each person who issued the opinions stored in the storage device 8.

[0019] The information processing device 2 is a computer terminal. In this embodiment, the analysis unit 5, the estimation unit 6, the dynamic dialogue generation unit 7, the storage device 8, and the management unit 9 are provided in the information processing device 2. Note that the analysis unit 5, the estimation unit 6, the dynamic dialogue generation unit 7, the storage device 8, and the management unit 9 may be provided in an external computer or server and connected to the information processing device 2 via an Internet line.

[0020] In this embodiment, the input means 3 is assumed to be a terminal that can be connected to the information processing device 2 via an Internet line, such as a smartphone or a computer terminal owned by a related party. Note that an input device such as a keyboard directly connected to the information processing device 2 may be used as the input means 3. Also, it may be used as an input means for acquiring opinions from the voice of a voice call.

[0021] The analysis unit 5 includes an emotion recognition engine 4, and analyzes the opinions input by the input means 3 using this emotion recognition engine 4 to score the emotional state.

[0022] In this analysis unit 5, the sentence is divided into words using morphological analysis (MeCab, SudachiPy, etc.), place names, facility names, organization names, etc. are identified by named entity recognition, and natural language processing is performed to remove words with little meaning (such as particles), while the spoken text is converted into a multidimensional vector using a large-scale language model (BERT, Japanese version of RoBERTa, etc.).

[0023] Subsequently, the emotion recognition engine 4 applies a pre-trained classification model using training data (past resident statements and assigned emotion labels) to output a probability (score) of 0.0 to 1.0 for each emotion category.

[0024] The estimation unit 6 estimates the consensus-building stage from the emotion scores analyzed by the analysis unit 5 and predetermined thresholds. Specifically, the estimation unit 6 calculates a negative emotion score by weighting the emotion scores for anger, distrust, and anxiety according to predetermined weights, based on the emotion scores calculated by the analysis unit 5. It then applies an exponential moving average to the calculated negative emotion score to smooth it over time and estimates the consensus-building stage based on predetermined thresholds and consecutive occurrence conditions. For example, the thresholds are set based on the following example: If negative emotion > 0.7, it is estimated to be the initial resistance stage, and if distrust < 0.4 and agreement > 0.6, it transitions to the next stage. If the same stage appears two or more times consecutively, it is estimated that the transition is confirmed. In this consensus-building stage, if it starts from the initial resistance stage, it transitions in the order of conditional acceptance, readiness for acceptance, and agreement.

[0025] In this embodiment, the determination of the consensus-building stage is set as follows: initial resistance if initial negative emotion ≥ 0.7, conditional acceptance if 0.40 ≤ negative emotion < 0.70 and 0.30 ≤ agreement < 0.60, readiness for acceptance if negative emotion < 0.40, distrust < 0.40, and agreement ≥ 0.50, and agreement if negative emotion < 0.25, distrust ≤ 0.25, and agreement ≥ 0.75. Here, the negative emotion score is calculated using a weighted average of (anger × 0.4) + (distrust × 0.35) + (anxiety × 0.25), and an exponential moving average (α = 0.4) is used to determine the consensus-building stage.

[0026] Stage transitions (transitions in the consensus-building stage) are determined when the following transition conditions are met consecutively for two consecutive times (three consecutive times for agreement only). A transition from initial resistance to conditional acceptance is determined when negative emotion ≤ 0.65 is met consecutively, and agreement ≥ 0.35 or openness ≥ 0.50 is met consecutively. Openness is calculated as Doubt / Concern Score × 0.5 + (1 - Anger Score) × 0.5. A transition from conditional acceptance to readiness for acceptance is determined when negative emotion < 0.40, distrust < 0.40, and agreement ≥ 0.50 are met consecutively. A transition from readiness for acceptance to agreement is determined when negative emotion < 0.25, distrust ≤ 0.25, and agreement ≥ 0.75 are met consecutively. Whether or not the transition conditions are met consecutively is determined by the value of the consecutive count counter stored in the management unit or memory device. The continuous numerical counter is incremented to 1 the first time a transition condition is met, and resets the next time the transition condition is not met.

[0027] Conversely, there are also cases where it is judged that the consensus-building stage has regressed. A regression in the consensus-building stage occurs when negative emotions ≥ 0.55 or distrust ≥ 0.50 occur consecutively while remaining in a higher stage, causing the group to move back one stage.

[0028] If none of these stage transition or reversal conditions are met, it is determined that the current consensus-building stage is being maintained.

[0029] By performing this type of processing, we can go beyond sentiment analysis based on instantaneous statements and introduce time-series processing such as exponential moving averages, thereby eliminating the speaker's transient emotions and enabling the estimation of a stable and objective consensus-building stage. The estimation unit 6 may automatically determine the consensus-building process, which consists of multiple stages: initial resistance, conditional acceptance, readiness for acceptance, and agreement, and generate a consensus-building stage visualization map that visualizes the results.

[0030] These thresholds should ideally be determined appropriately based on historical data and statistical analysis of pilot operations (ROC curves, F-value maximization, etc.). Furthermore, it is preferable to determine whether the thresholds are met using an exponential moving average (EMA, α=0.4) rather than the instantaneous value of the utterance unit.

[0031] By incorporating such consensus-building algorithms (weighted average, EMA, threshold determination) and linking them with attribute estimation, the progress of consensus building can be quantified and visualized, enabling support for consensus building with transparency and accountability.

[0032] Furthermore, the estimation unit 6 generates a consensus-building stage visualization map that visualizes the estimated consensus-building stages.

[0033] The dynamic dialogue generation unit 7 automatically generates dialogue guidance templates according to the estimated emotional state and consensus-building stage, supporting two-way dialogue with residents or stakeholders. The dynamic dialogue generation unit 7 only needs to generate dynamic dialogue according to the consensus-building stage estimated by the estimation unit 6, but in this embodiment, it also refers to the emotional score to automatically generate dynamic conversations.

[0034] Furthermore, in this embodiment, the dynamic dialogue generation unit 7 automatically adjusts the generated dialogue content according to the direction and magnitude of the multidimensional vector calculated by the emotion recognition engine 4.

[0035] In this embodiment, the management unit 9 includes an attribute estimation module 10 that analyzes the speaker's speech history and behavioral patterns and estimates attributes such as reasons for opposition, areas of interest, and influencing factors. This attribute estimation module 10 provides the analyzed attributes to the dynamic dialogue generation unit 7, which then automatically generates dialogue content by further referring to these attributes.

[0036] Specifically, the attribute estimation module 10 analyzes the sentiment score trends, keyword frequency, affiliated organizations, family structure, educational background, and relationship with the timing of relevant news reports in the speech history of the person who inputs the opinion, and classifies their areas of interest. The dynamic dialogue generation unit 7 then adjusts the information presented and the style of expression in the dialogue content based on the estimated areas of interest and social background. In attribute estimation, in addition to basic attributes such as age group and family structure, speech tendencies and external factors are analyzed in an integrated manner and reflected in the dynamic dialogue generation, thereby providing an explanation and persuasion process optimized for the speaker.

[0037] This dynamic dialogue generation unit 7 outputs the automatically generated dialogue content as text data and audio data, and also transmits it to the terminal of the person who expressed the opinion. This transmission also includes making a call using audio data.

[0038] The storage device 8 stores the emotional state of each person who expressed an opinion, the stage of consensus building, and the dialogue history automatically generated by the dynamic dialogue generation unit 7.

[0039] The management unit 9 centrally manages the emotional state of each person who expressed an opinion, the consensus-building stage, and the dialogue history automatically generated by the dynamic dialogue generation unit 7, all of which are stored in the memory device 8. Because the management unit 9 centrally manages this information, it can use the attribute estimation module 10 to update attributes such as reasons for opposition, areas of interest, and influencing factors, and provide them to the dynamic dialogue generation unit 7 as appropriate, or store the attributes in the memory device 8.

[0040] Furthermore, the management unit 9 automatically records the content of the dialogue, the history of emotional changes, and the progress of the consensus-building stage, and preserves them as evidence to fulfill accountability.

[0041] The consensus-building support method 11 of the present invention is mainly performed using the consensus-building support system 1, and as shown in Figure 2, consists of: an opinion acquisition step 12 for acquiring the opinions of stakeholders; an emotion analysis step 13 for analyzing the opinions acquired in the opinion acquisition step 12 using an emotion recognition engine 4 and calculating the emotional state as an emotion score; an agreement-building stage estimation step 14 for estimating the agreement-building stage from the emotion score analyzed in the emotion analysis step 13; a dynamic dialogue generation step 15 for automatically generating dialogue content based on the agreement-building stage estimated in the agreement-building stage estimation step 14 for the opinions acquired in the opinion acquisition step 12, and transmitting the dialogue content to the person who expressed the opinion; and a management step 16 for storing and centrally managing the emotional state of each person who expressed the opinion, the agreement-building stage, and the history of the dialogue content automatically generated in the dynamic dialogue generation step 15 in a storage device.

[0042] In opinion acquisition step 12, opinions are acquired using input means 3, such as smartphones and computer terminals, owned by the relevant parties. These acquired opinions are stored in a storage device 8 connected to the information processing device 2. In this embodiment, when a user accesses the information processing device 2 via the internet using a terminal such as a smartphone, enters login information as needed, and inputs their opinion via the smartphone, the opinion is automatically stored in the storage device 8 of the information processing device 2. It is desirable that the input means 3 be equipped with a conversion means that converts the opinion, if input is given via voice, into text data using voice recognition software or the like.

[0043] In the emotion analysis step 13, the opinions entered in the opinion acquisition step 12 are analyzed by the analysis unit 5 provided in the information processing device 2. Since this emotion analysis step 13 is performed each time an opinion is entered, the opinions of the same person and the emotion scores analyzed in the emotion analysis step 13 are centrally managed in the management step 16 so that they can be compared as time-series data for the same person.

[0044] In the analysis unit 5, the input opinion text data is first divided into words and other elements using morphological analysis such as MeCab / Sudachi. For example, if the opinion "This plan is very dangerous and I am worried about the children" is obtained in opinion acquisition step 12, it is divided into "This / plan / is / very / dangerous / and / children / are / worried / ." After that, facility names, regional names, and stakeholder names are extracted, and stop words and words that do not contribute to sentiment determination, such as "is," "is," and "is," are removed.

[0045] Subsequently, a large-scale language model (BERT, Japanese version RoBERTa, etc.) is used to convert the context of the utterance into a multidimensional vector (context vector). In addition, auxiliary features such as sentiment dictionary scores (positive / negative), the number of times emphatic expressions ("very," "extremely") are used, and the presence or absence of negative expressions ("~not") are calculated.

[0046] The opinions, which have been transformed into multidimensional vectors in this way, are classified into multiple classes of emotions (joy, anger, sadness, happiness, anxiety, distrust, agreement, doubt, misinformation) using the nn.softmax function. The emotion recognition engine 4 then refers to the auxiliary features and calculates an emotion score for each opinion using the softmax function and training data, and aggregates each emotion score. Furthermore, among these emotion scores, the scores for negative emotions such as anger, anxiety, and distrust (negative emotion scores) are weighted and calculated as a weighted average. For example, the negative emotion score is calculated by weighting it as follows: Negative emotion score = (anger × 0.4) + (distrust × 0.35) + (anxiety × 0.25).

[0047] After calculating the emotional state score (emotion score), the emotional score (0.0 to 1.0) for each emotional category is output in JSON format or similar, stored in the memory device 8, and the process proceeds to the consensus building stage estimation step 14.

[0048] In the consensus-building stage estimation step 14, the current consensus stage (consensus-building stage) of stakeholders such as residents is estimated by comparing the emotion score calculated in the emotion analysis step 13 with a set threshold. For example, if the negative emotion score is greater than 0.7, it is determined to be initial resistance.

[0049] This consensus-building stage estimation step 14 is performed each time an opinion is entered, and the consensus-building stage is estimated each time. If distrust < 0.4 and agreement > 0.6, the system proceeds to the next stage. If the same stage occurs two or more times in a row, the estimation unit 6 determines that the transition to the consensus-building stage is confirmed.

[0050] In the dynamic dialogue generation step 15, dialogue content is automatically generated based on the opinion obtained in the opinion acquisition step 12 and the consensus building stage estimated in the consensus building stage estimation step 14. It is desirable that the automatically generated dialogue content be automatically adjusted not only according to the consensus building stage but also according to the direction and magnitude of the multidimensional vector calculated in the emotion analysis step 13. Furthermore, in this embodiment, the attributes of the person who input the opinion in the opinion acquisition step 12 are estimated using the attribute estimation module 10, and the dialogue content is automatically generated by further referencing these attributes.

[0051] Attribute estimation is used to evaluate residents' social backgrounds and living situations from multiple perspectives, and to optimize the interpretation of their statements and consensus-building strategies.

[0052] The attribute information used includes age group, region, affiliated organizations (neighborhood associations, opposition groups, etc.), family structure (single, married, families with children, elderly family members living together), educational background (high school graduate, junior college graduate, university graduate, postgraduate, etc.), past speaking trends (emotion score trends, keyword frequency), and media exposure such as related news and social media posts.

[0053] Specifically, the system uses information entered by those who submit their opinions to obtain details such as age group, region, affiliated organizations (neighborhood associations, opposition organizations, etc.), family structure (single, married, families with children, elderly family members living together), and educational background (high school graduate, junior college graduate, university graduate, postgraduate, etc.). It also extracts areas of interest from comments (e.g., "environment," "safety," "economic burden") and classifies them into categories such as health concerns, economic burdens, and procedural distrust. Furthermore, it estimates the social background assessment, including the severity of the areas of interest and the level of information comprehension, based on family structure and educational background. Finally, it analyzes the relationship (influencing factors) with geographical proximity, organizational affiliation, and the timing of news reporting.

[0054] Based on the attributes estimated from this information, the system automatically generates dialogue content with the person who submitted the opinion. For the dialogue content, a base dialogue is selected from templates categorized by consensus-building stage and emotion score. Unique information, such as attributes, concerns, region names, and summaries of past statements, is then added to this template. Furthermore, the dialogue style is adjusted (gentle / scientific / concise) considering the attributes of the person who submitted the opinion before outputting the dialogue. This outputted dialogue content is then sent to the person who submitted the opinion via text, email, voice message, etc.

[0055] In management step 16, the dialogue history automatically generated in the consensus building stage and dynamic dialogue generation step 15 is automatically stored in a storage device and centrally managed.

[0056] In the embodiments of this invention, Mecab, BERT, RoBERTa, and softmax classification are used, but these are merely examples, and other natural language processing engines may be used. [Industrial applicability]

[0057] This invention is used in industries that require consensus building among multiple parties. [Explanation of Symbols]

[0058] 1: Consensus building support system, 2: Information processing device, 3: Input method, 4: Emotion recognition engine, 5: Analysis section, 6: Estimation section, 7: Dynamic dialogue generation unit, 8: Storage device, 9: Management Department, 10: Attribute Estimation Module, 11: Methods for supporting consensus building, 12: Steps for obtaining opinions, 13: Sentiment analysis step, 14: Consensus building stage estimation step, 15: Dynamic dialogue generation step, 16: Management step.

Claims

1. The system comprises an information processing device, an input means for inputting the opinions of relevant parties into the information processing device, an analysis unit that analyzes the opinions input by the input means using an emotion recognition engine and calculates the emotional state as an emotion score, an estimation unit that estimates the consensus formation stage from the emotion score analyzed by the analysis unit, a dynamic dialogue generation unit that automatically generates dialogue content based on the consensus formation stage estimated by the estimation unit for the opinions and transmits the dialogue content to the person who expressed the opinion, a storage device that stores the history of the emotional state, consensus formation stage, and dialogue content automatically generated by the dynamic dialogue generation unit for each person who expressed the opinion, and a management unit that manages the history of the emotional state, consensus formation stage, and dialogue content automatically generated by the dynamic dialogue generation unit stored in the storage device. The analysis unit, the estimation unit, the dynamic dialogue generation unit, the storage device, and the management unit are provided in the information processing device or connected to the information processing device via an internet connection. The estimation unit calculates a negative emotion score by weighting the emotion scores for anger, distrust, and anxiety according to predetermined weights, based on the emotion scores calculated by the analysis unit, applies an exponential moving average to the calculated negative emotion score to smooth it over time, and estimates the consensus-building stage based on predetermined thresholds and consecutive occurrence conditions.

2. The consensus building support system according to claim 1, characterized in that the emotion recognition engine of the analysis unit uses natural language processing and a large-scale language model to analyze the context of the opinion and calculate multiple emotion categories as multidimensional vectors, and the dynamic dialogue generation unit automatically adjusts the dialogue content to be generated according to the direction and magnitude of the multidimensional vectors calculated by the emotion recognition engine.

3. The consensus building support system according to claim 1 or 2, characterized in that the management unit includes an attribute estimation module that analyzes the speaker's speech history and behavioral patterns and estimates attributes, provides the attributes analyzed by the attribute estimation module to the dynamic dialogue generation unit, and the dynamic dialogue generation unit further references the attributes to automatically generate dialogue content.

4. The consensus building support system according to claim 3, characterized in that the attribute estimation module analyzes the relationship between the sentiment score changes, keyword frequency, affiliated organizations, family structure, educational background, and relevant news reporting period of the person who input the opinion, classifies the area of ​​interest, and the dynamic dialogue generation unit adjusts the information presented and the style of expression of the dialogue content based on the estimated area of ​​interest and social background.

5. The consensus-building support method comprises: an opinion acquisition step of acquiring opinions from stakeholders; an emotion analysis step of analyzing the opinions acquired in the opinion acquisition step using an emotion recognition engine and calculating the emotional state as an emotion score; an agreement stage estimation step of estimating the agreement stage from the emotion score analyzed in the emotion analysis step; a dynamic dialogue generation step of automatically generating dialogue content and sending the dialogue content to the person who expressed the opinion based on the agreement stage estimated in the agreement stage estimation step for the opinions acquired in the opinion acquisition step; and a management step of storing and centrally managing the history of the emotional state for each person who expressed the opinion, the agreement stage, and the dialogue content automatically generated in the dynamic dialogue generation step in a storage device. In the agreement stage estimation step, a negative emotion score is calculated by weighting the emotional scores of anger, distrust, and anxiety with predetermined weights based on the emotion score, and the agreement stage is estimated based on predetermined thresholds and consecutive occurrence conditions by applying an exponential moving average to the calculated negative emotion score to smooth it over time.

6. The consensus building support method according to claim 5, characterized in that in the emotion analysis step, the emotion recognition engine uses natural language processing and a large-scale language model to analyze the context of the opinion and calculate multiple emotion categories as multidimensional vectors, and in the dynamic dialogue generation step, the dialogue content to be generated is automatically adjusted according to the direction and magnitude of the multidimensional vectors calculated by the emotion recognition engine.

7. The consensus-building support method according to claim 5 or 6, characterized in that the opinion acquisition step estimates the attributes of the person who input the opinion using an attribute estimation module, and further references the attributes to automatically generate dialogue content, the attribute estimation module analyzes the sentiment score changes, keyword frequency, affiliated organizations, family structure, educational background, and relationship with the period of relevant news coverage of the person who input the opinion to classify areas of interest, and the dynamic dialogue generation step adjusts the information presented and the style of expression of the dialogue content based on the estimated areas of interest and social background.

Images