Action suggestion system, information processing device, information processing method, and program

Abstract

This system provides the ability to suggest actions that users may not be aware of, based on their activity history. [Solution] A system comprising a control unit and a storage unit is provided, wherein the control unit includes an acquisition module that acquires historical information relating to the user's activities, a generation module that divides the received historical information into morphological units and generates historical segmentation information combined with time-series information, an assignment module that extracts characteristic words from the historical segmentation information and assigns corresponding classification labels, a calculation module that calculates the importance of the assigned classification labels, and a suggestion module that outputs suggestion information regarding the next action the user should take based on predetermined evaluation criteria for the classification labels and the importance.

Description

Technical Field

[0001] Embodiments of the present disclosure relate to a behavior recommendation system, an information processing apparatus, an information processing method, and a program.

Background Art

[0002] In recent years, systems have been developed that acquire information related to a user's activity history and analyze the information to assist and recommend the user's next behavior. For example, Patent Document 1 discloses a system that acquires a user's schedule information and determines the date and time of a recommended action.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in the above-described system, since actions are recommended based on the schedule planned by the user, it is not possible to propose actions that the user does not intend.

[0005] The problem to be solved by the present invention is to provide a system capable of proposing actions that the user does not intend based on the user's activity history.

Means for Solving the Problems

[0006] As one aspect of the present disclosure, an action suggestion system is provided, comprising a control unit and a storage unit, wherein the control unit includes an acquisition module for acquiring historical information relating to the user's activities, a generation module for dividing the received historical information into morphological units and generating historical segmentation information combined with time-series information, an assignment module for extracting characteristic words from the historical segmentation information and assigning corresponding classification labels, a calculation module for calculating the importance of the assigned classification labels, and a suggestion module for outputting suggestion information regarding the next action the user should take based on predetermined evaluation criteria for the classification labels and the importance. [Brief explanation of the drawing]

[0007] [Figure 1] This is a schematic diagram showing the overall configuration of the action suggestion system 1 according to the embodiment. [Figure 2] This is a block diagram showing the functional configuration of server 200 in the action suggestion system 1. [Figure 3] This is a schematic diagram showing an example of the database configuration of the memory unit 220. [Figure 4] This is a flowchart showing the overall processing flow of the Action Suggestion System 1. [Figure 5] This is a flowchart showing the processing flow of the history splitting process (S100). [Figure 6] This is a flowchart showing the processing flow of the importance calculation process (S200). [Figure 7] This is a flowchart showing the processing flow of the proposal output process (S300). [Figure 8] This figure shows an example of a daily report input screen 51 displayed on the user terminal 100. [Figure 9] This figure shows an example of a proposal output screen 52 displayed on the user terminal 100. [Figure 10] This is a block diagram showing the hardware configuration of computer 90, which is an information processing device. [Modes for carrying out the invention]

[0008] <1. Embodiments> The following describes the action suggestion system 1 as an embodiment of this design, with reference to the drawings. In this specification and in each drawing, elements already described are denoted by the same reference numerals, and redundant explanations are omitted as appropriate. In this embodiment, "module" refers to a concept that includes processing functions realized by hardware resources, or information processing mechanisms that realize such functions in software.

[0009] (1.1. Overview of the Action Suggestion System) As shown in Figure 2, the action suggestion system 1 comprises a user terminal 100 operated by the user, a server 200 that performs the main processing related to action suggestions, and an external AI server 300 equipped with Generative Artificial Intelligence (Generative Artificial Intelligence). These are interconnected and can communicate with each other via a network N.

[0010] The user terminal 100 is an information processing device for users to input information about their daily activity history (for example, daily reports on work content, etc.) and receive suggestion information from the server 200. Specifically, it inputs history information in natural language through a work report input screen and outputs analysis results and action suggestions from the server 200. The user terminal 100 can be configured as a tablet terminal, smartphone, or PC terminal, etc.

[0011] Server 200 is an information processing device that serves as the central hub for action suggestion processing, and comprises a control unit 210 and a storage unit 220. The control unit 210 includes an acquisition module 211, a generation module 212, an assignment module 213, a calculation module 214, a suggestion module 215, and an evaluation module 216. These modules acquire historical information about the user's activities, generate segmented history information divided into morphological units, extract feature words and assign tags, calculate the importance of the tags, and then generate action suggestion information based on predetermined evaluation criteria. The evaluation module 216 calculates the reliability of the suggestion information using statistical methods.

[0012] The memory unit 220 stores a history segmentation database 221, an importance database 222, an evaluation criteria database 223, and a proposal history database 224. The history segmentation database 221 holds history segmentation information and time-series information segmented into morphological units. The importance database 222 holds tags as classification labels assigned to the history segmentation information and scores as importance assigned to those tags. The evaluation criteria database 223 holds criteria for proposal judgment regarding the transition patterns and probability distributions of tags. The proposal history database 224 holds generated proposal information and acceptance / rejection results.

[0013] The external AI server 300 is an information processing device equipped with a generative AI model, which performs processing to assist in natural language processing and action suggestion generation. For example, the external AI server 300 includes a generative engine using a large language model (LLM). Server 200 works in conjunction with the external AI server 300 to supplement and improve the accuracy of suggested information.

[0014] For example, the external AI server 300 may be a general-purpose generative AI model such as ChatGPT or Gemini, or it may be a dedicated model specialized for a specific business area. The external AI server 300 has the function of generating text output of possible next actions and recommended tasks based on the time-series changes of input history information and tags, thereby realizing flexible and highly accurate action suggestion support.

[0015] (1.2. Functional configuration of Server 200) Referring to FIG. 2, the functional configuration of server 200 will be described. Server 200 includes a control unit 210 and a storage unit 220. The control unit 210 includes an acquisition module 211, a generation module 212, an assignment module 213, a calculation module 214, a proposal module 215, and an evaluation module 216. These modules function by the control unit 210 reading the programs and data stored in the storage unit 220, and by delivering the necessary data to the external AI server 300, the historical information regarding the user's activities is analyzed to generate action proposals.

[0016] The acquisition module 211 acquires historical information regarding activities transmitted from the user terminal 100. The historical information includes natural language data such as business daily reports, schedule records, task inputs, and speech recognition results. The acquisition module 211 converts this information into a predetermined format.

[0017] The generation module 212 executes a process of delivering the acquired historical information to the external AI server 300, splitting it into words or phrases by morphological analysis, and attaching time-series information to generate historical split information. The generation module 212 is configured to assign an order key to each piece of historical information and record the split information in chronological order, so as to efficiently perform subsequent feature word extraction and tagging processes.

[0018] Based on the generated historical split information, the assignment module 213 causes the external AI server to execute a process of extracting feature words and attaching corresponding classification labels (tags) to each of them. Here, the extraction of feature words is realized using co-occurrence degree, TF-IDF value, or embedding representation by a neural network. As a result, it is configured to quantitatively grasp the contextual features within the historical information.

[0019] The calculation module 214 causes an external AI server to calculate the importance of the tags assigned by the assignment module 213. The importance may be defined as the probability distribution of occurrence for each tag, or as the posterior probability obtained by Bayesian estimation. As an example, the calculation module 214 calculates the importance of the tag probability distribution corresponding to each ordinal key and stores it in the importance database 222.

[0020] The suggestion module 215 generates suggestion information for the user's next action, using the probability distribution of tags obtained by the calculation module 214 and the evaluation criteria stored in the evaluation criteria database 223. Here, the evaluation criteria depend only on the probability distribution of tags and the transition patterns along the sequential key, and have a general configuration that does not depend on the external environment or individual attributes. The generated suggestion information is stored in the suggestion history database 224.

[0021] The evaluation module 216 performs a process to evaluate the reliability of the proposal information generated by the proposal module 215 using statistical methods. Specifically, it numerically calculates the reliability of the proposal results using methods such as variance estimation of the score distribution of the proposal information and bootstrap testing. In addition, it obtains whether the user actually adopted the proposal after it was submitted and dynamically updates the reliability based on that acceptance / rejection result.

[0022] The storage unit 220 is a storage device that stores various types of data in the server 200, and includes a history division database 221, an importance database 222, an evaluation criteria database 223, and a proposal history database 224. These databases function as basic information for analyzing historical information and generating proposals.

[0023] (1.3. Data Structure) Referring to Figure 3, the data structure stored in the storage unit 220 will be explained. Note that the data structure described below is merely an example, and the items included in each database may be added, deleted, or merged depending on the operating mode of the system.

[0024] As shown in Figure 3, the storage unit 220 includes, as an example, a history division database 221, an importance database 222, an evaluation criteria database 223, and a proposal history database 224.

[0025] The history segmentation database 221 is a database that stores historical information about user activities, segmented into morphological units. For example, the history segmentation database 221 includes the fields "Case ID," "Time," "User Identifier (Who)," "Action Target (What)," and "Action Result (What Happened)." The "Case ID" field is an identifier used to uniquely identify records in the history segmentation database 221 and functions as a primary key. This facilitates data storage and retrieval on a user-by-user or time-series basis.

[0026] The importance database 222 is a database that stores tags extracted based on historical information recorded in the history segmentation database 221, along with their scores representing their importance. The importance database 222, for example, includes the fields "Case ID," "Tags," and "Score (Importance)." The score is defined as a probability distribution or normalized value and is updated by the calculation module 214.

[0027] The evaluation criteria database 223 is a database that stores criterion information for proposing actions based on the probability distribution and transition patterns of tags. For example, the evaluation criteria database 223 has the items "Criterion ID", "Proposal Type", "Proficiency", "Action Content", and "Evaluation Score". The item "Criterion ID" is the primary key for uniquely identifying each criterion record. The "Evaluation Score" is defined as a value calculated based on the pattern of tag transitions along the probability distribution and ordinal key. The evaluation criteria database 223 registers numerical criteria for evaluating action sequences (the changes in the time series of actions), such as the consistency of the action sequence, the degree of variability (entropy reduction), and the probability of tag transitions occurring.

[0028] The proposal history database 224 is a database that stores proposal information generated by the proposal module 215 and its adoption results. The proposal history database 224 includes, for example, the fields "Proposal ID," "Case ID," "Proposal Content," "Reliability," and "Adoption Status." The "Proposal ID" field is the primary key of the proposal history database 224, and the adoption status information is updated by the evaluation module 216. The data stored in the proposal history database 224 is used for improving proposal information and evaluating its reliability.

[0029] In this way, the memory unit 220 can achieve highly accurate suggestion generation based on the user's past behavior and continuous improvement of accuracy by interrelating and storing multiple data structures such as history information, tags, evaluation criteria, and suggestion history.

[0030] (1.4. Processing Procedure) The processing flow in the action suggestion system 1 will be explained by referring to Figures 4 to 7. First, the overall flow will be explained by referring to Figure 4.

[0031] (1.4.1. Overall Flow) Figure 4 is a flowchart showing the overall processing flow in the Action Suggestion System 1. Based on the activity history information acquired from the user terminal 100, the Action Suggestion System 1 generates action suggestion information by having the server 200 and the external AI server 300 collaborate to perform a series of data analysis processes. As shown in Figure 4, the Action Suggestion System 1 mainly executes the following three main steps in order: the history segmentation process (step S100), the importance calculation process (step S200), and the suggestion output process (step S300).

[0032] In the history segmentation process (step S100), first, information regarding activity history (e.g., daily work reports, task records, chat history, speech logs, etc.) entered via the user terminal 100 is sent to the server 200. The acquisition module 211 of the server 200 receives this history information, and the generation module 212 passes this information and prompts (instructions) to the external AI server 300, thereby generating history segmentation information divided into morphological units. Furthermore, metadata such as input date and time, document type, and inputter ID is assigned to each segmentation unit, and a sequential key is assigned, so that the entire history is systematically managed along the time axis. This preprocessing allows for high-precision analysis of contextual and chronological relationships when extracting feature words and calculating importance in subsequent steps. Details of this process will be described later with reference to Figure 5.

[0033] In the importance calculation process (step S200), tags are used as classification labels for the history segmentation information generated in the history segmentation process (S100) to calculate a score representing the importance of each tag. The calculation module 214 statistically analyzes the frequency of tag occurrences, co-occurrence relationships, and transition trends, and quantifies the extent to which each tag characterizes the user's behavioral tendencies as a probability distribution. At this time, the score is not merely a frequency index, but is calculated considering weighting based on the relationship structure between tags and the order key of their occurrence positions. This makes it possible to extract potential trends and periodicities from past behavioral history and estimate the characteristics of the next likely behavior. Details of this process will be described later with reference to Figure 6.

[0034] In the suggestion output process (step S300), the system estimates the next action the user should take based on the calculated tag importance and the criteria information stored in the evaluation criteria database 223, and outputs it as suggestion information. The suggestion module 215 extracts the time-series changes of tags that are highly important and show significant changes in transition probability, and derives the optimal action pattern by comparing them with the evaluation criteria. The suggestion information includes elements such as the action content, implementation timing, and reasons for recommendation, and is presented to the user terminal 100. Furthermore, the evaluation module 216 performs a reliability evaluation of the suggestion results using statistical methods. Specifically, it calculates the likelihood of each suggestion by referring to the distribution of suggestion scores and past adoption history, and registers the reliability in the suggestion history database 224. This allows for continuous feedback of the acceptance or rejection results after the suggestion, and dynamically improves the accuracy of the action suggestions. Details of this process will be described later with reference to Figure 7.

[0035] (1.4.2. History Partitioning Process) Refer to Figure 5 for a detailed explanation of the history segmentation process (S100). Figure 5 is a flowchart showing the processing flow of the history segmentation process (S100). The history segmentation process (S100) is a process in which historical information about the user's activities is divided into morphological units using natural language processing, and history segmentation information is generated and stored chronologically.

[0036] In S110, the system acquires historical information about the user's activities entered on the daily report input screen 51 of the user terminal 100. This historical information includes work records, spoken content, input documents, location information, and operation logs of various applications. The daily report input screen 51 shown in Figure 7 is an example screen for users to input business indicators such as the number of contracts, the number of business negotiations, and the number of visits, as well as activity history and impressions. The entered information is sent to the server 200 as historical information about the activities. The acquisition module 211 acquires this information, normalizes it into a predetermined format, and passes it to the generation module 212 for subsequent processing.

[0037] In S120, the generation module 212 passes the historical information acquired by the acquisition module 211 to the external AI server 300 along with a prompt, where it is divided into morphological units by natural language processing and historical segmentation information is generated. Morphological analysis uses a combination of dictionary-based part-of-speech segmentation and machine learning models to segment words and phrases according to the context.

[0038] Furthermore, metadata such as input date and time, source (application type, etc.), and inputter ID are assigned to each history unit, and a sequential key is assigned to clarify the temporal continuity between history entries. As a result, history information is organized not merely as document data, but as structured data that allows for a quantitative understanding of the flow of actions, improving the accuracy of estimations in subsequent processing, such as importance calculation and action suggestions.

[0039] In S130, the generation module 212 stores the history partition information, to which an order key order_k has been assigned according to a predetermined rule, in the history partition database 221 of the storage unit 220. Here, the generation of the order key is performed based on a predetermined rule for uniquely identifying the order and relationships of history occurrences. As an example of the predetermined rule, a configuration may be used in which a monotonically increasing order key is assigned by a deterministic sorting rule that does not depend on the physical storage order or arrival order. With such a configuration, even if there is a disturbance in the arrival order of history data due to communication delays or parallel processing, the history partition information can be registered while maintaining temporal consistency. In the storage unit 220, the history partition information is classified and managed for each user ID and used as input data for subsequent tagging processing and probability distribution calculation processing.

[0040] Through the above processing, the history segmentation process (S100) structures the activity history described in natural language at the morphological level and organizes it as history segmentation information that maintains chronological relationships. Performing this preprocessing improves the accuracy of the analysis in the subsequent importance calculation process (S200) and enables highly reliable action suggestions.

[0041] (1.4.3. Flow of the Importance Calculation Process) Refer to Figure 6 to explain the detailed processing flow of the importance calculation process (S200). Figure 6 is a flowchart showing the processing flow of the importance calculation process (S200). In the importance calculation process (S200), tags are assigned as classification labels to the history segmentation information, the occurrence trend of each tag is analyzed, and importance is calculated. This allows for the extraction of information that characterizes the user's behavioral trends and the generation of data that forms the basis for estimating the next action to take.

[0042] In S210, the assignment module 213 retrieves the history segmentation information generated in the history segmentation process (S100) from the history segmentation database 221. The retrieved history segmentation information contains information about the user's activity history, which has been segmented and stored while retaining chronological information.

[0043] In S220, the process of extracting characteristic words from the acquired history segmentation information is executed. The assignment module 213 passes the history segmentation information to the external AI server 300, which then performs characteristic word extraction processing based on natural language analysis. This extracts contextually important words and phrases from each history unit, such as keywords like "meeting," "business negotiation," "report," "input," and "movement." The analysis by the external AI server 300 takes into account word frequency, dependency structure, and co-occurrence relationships, and the extraction results are returned to the server 200.

[0044] In S230, the tagging module 213 tags the history segmentation information based on the extracted characteristic words. The tagging process is also executed by passing input data and prompts to the external AI server 300. Tags are classification labels that represent the user's activities and correspond to action types such as "meeting," "travel," "document creation," and "browsing." Based on the results returned from the external AI server 300, the tagging module 213 assigns appropriate tags to each history unit. In this way, tag data that semantically classifies the user's actions is generated.

[0045] In S240, the calculation module 214 passes input data and prompts to the external AI server 300, calculates the probability distribution of tags assigned to the history segmentation information as importance, and stores it in the storage unit 220 along with the order key. Specifically, the calculation module 214 extracts statistical features such as the frequency of occurrence of each tag included in the history segmentation information, the co-occurrence relationship with the preceding and succeeding tags, and the number of tag transitions, and normalizes these to generate a probability distribution p_z(k) on the tag set Z for tag k included in the history segmentation information. In addition, the probability value p(z|k) of a specific tag z in the generated probability distribution may be calculated as the importance of that tag.

[0046] Furthermore, the external AI server 300 may apply a probabilistic model (e.g., a Markov model, Bayesian network, or neural network) to these statistical features as input and output an importance distribution corresponding to each tag. The calculation module 214 may sort the obtained probability distributions in ascending order of the order key order_k and store them in the importance database 222 as {(order_k, p_z(k))}. This systematically stores the tendency of label appearance and chainlines in each history unit as a probabilistic indicator, enabling the subsequent suggestion output process (S300) to prioritize suggesting actions that consider tags with high importance. In addition, the probability distribution column allows for a statistical understanding of user behavior trends, turning points, and repetitive patterns independently of time series, thereby improving the objectivity and reproducibility of suggestions.

[0047] Thus, in the importance calculation process (S200), tags are assigned based on characteristic words extracted from the history segmentation information, and the importance is statistically calculated using the frequency of occurrence, co-occurrence relationships, and transition probabilities of each tag. The calculation module 214 may quantitatively evaluate the stability and consistency of behavioral tendencies by analyzing the shape of the probability distribution p_z(k) (e.g., variance value and entropy value). With this configuration, it is possible to probabilistically model the semantic relationships and flow of behavior in the user's behavioral history and generate objective evaluation indicators that form the basis of behavioral suggestions. As a result, it is possible to generate suggestions that reflect the user's intentions and behavioral structure with higher accuracy compared to conventional methods based on simple frequency aggregation.

[0048] (1.4.4. Proposal Output Process Flow) Refer to Figure 7 to explain the detailed processing flow of the proposal output process (S300). Figure 7 is a flowchart showing the processing flow of the suggestion output process (S300). In the suggestion output process, the importance and probability distribution of tags output from the importance calculation process (S200) are obtained, and suggestion information regarding user behavior is generated based on predetermined evaluation criteria. Furthermore, the reliability of the generated suggestion information is evaluated, and a decision is made as to whether or not to output it based on the result.

[0049] In S310, the suggestion module 215 retrieves the importance and probability distribution data of the tags obtained in the importance calculation process (S200) from the importance database 222. Based on this data, the suggestion module 215 analyzes the overall trends of the user's behavior history and extracts the information necessary to generate suggestion candidates.

[0050] In S320, the proposal module 215 retrieves evaluation criteria stored in the evaluation criteria database 223. The evaluation criteria are a set of indicators that are commonly referenced for behavior estimation and proposal decision-making. For example, the evaluation criteria correlate with the probability of tag occurrence, co-occurrence relationships, and transition trends over time, and do not depend on other external factors (such as user attributes or environmental information). The proposal module 215 retains the retrieved evaluation criteria as the basis for scoring in the subsequent proposal generation process (step S330), and prepares to make a probabilistically valid proposal selection.

[0051] In S330, the suggestion module 215 generates suggestion information about the user's next action based on evaluation criteria. This process is performed by the suggestion module 215 passing the necessary input data and prompts to the external AI server 300. The external AI server 300 generates action suggestions by performing statistical optimization processing based on the information from the input importance database 222 and evaluation criteria. The suggestion module 215 refers to the evaluation criteria and extracts actions with high probability of consistency from among multiple action candidates, and stores them as suggestion candidates.

[0052] Specifically, proposed module 215 calculates the evaluation of a tag sequence using an evaluation function M[{p_z}] based solely on the probability distribution sequence {p_z(k)} or its transitions (tag changes). Here, the order key order_k is used only for sorting the sequence and is not included as an explanatory variable in the evaluation function. The evaluation function M[{p_z}] is defined, for example, as follows: First, the entropy (spread of the distribution) is calculated for the probability distribution p_z(t) at each time t, and a lower mean value (i.e., the more the distribution is concentrated on a particular tag) is given a higher evaluation. In addition, the rate of change between consecutive probability distributions (e.g., cosine similarity or KL divergence) is calculated, and sequences with slower changes are evaluated as having higher behavioral consistency. These values ​​are summed up with pre-set weights to calculate the score for the entire sequence as M[{p_z}].

[0053] The proposal module 215 extracts tags corresponding to action sequences whose scores, calculated by this evaluation function, exceed a predetermined threshold, and outputs the action content indicated by those tags as proposal information. The generated proposal information includes the content of the recommended action, the recommended timing for execution, the supporting tag sequence, and the calculated confidence level. As a result, the proposal results are independent of the time scale and input format, and consistent suggestions based on the structural characteristics of the action sequence itself can be provided.

[0054] In S340, the evaluation module 216 evaluates the reliability of the proposed information before output using statistical methods. This process is performed by the evaluation module 216 passing the necessary input data and prompts to the external AI server 300. The external AI server 300 receives multiple features constituting the proposed information as input variables, namely the variance of the tag probability distribution p_z(k), the entropy value based on the distribution shape, and similarity indices between consecutive distributions (e.g., cosine similarity or KL divergence), and analyzes these using statistical methods. Specifically, it calculates a confidence score (a continuous value between 0 and 1) based on the input variables using methods such as Bayesian estimation, logistic regression, and weighted average. For example, if the variance and entropy are low, and the similarity between adjacent distributions is high, it is judged to be stable as a consistent behavioral sequence, and the confidence level is set high. On the other hand, if these values ​​exceed a threshold, the confidence level is evaluated low, as the uncertainty of predicting the behavior is high. In this way, by statistically evaluating the reliability of the proposed information before output, a certain level of proposal accuracy can be ensured even when there is little historical information.

[0055] Furthermore, the evaluation module 216 may determine whether past proposal information has been adopted by the user and adjust the reliability of the generated proposal information based on the result. For example, for adopted proposal information, the reliability of proposals with similar feature structures may be adjusted upward, and for unadopted proposals, the reliability may be adjusted downward by updating the weights. This allows for a dynamic combination of history-dependent learning and history-independent statistical evaluation. The calculated reliability score may be recorded in the proposal history database 224 and used for output control by the proposal module 215. For example, proposal information with a reliability below a predetermined threshold may be excluded from output to the user terminal, and only proposal information above the threshold may be presented. This allows the action proposal system 1 to dynamically select and present only statistically valid and reliable proposals, and to continuously optimize proposal accuracy by reflecting the learning results.

[0056] In S350, the proposal module 215 outputs the generated proposal information that has been evaluated as having a confidence level exceeding a predetermined threshold to the proposal output screen 52 of the user terminal 100. As shown in Figure 9, the proposal output screen 52 displays the recommended action, recommended timing for execution, supporting tag sequence, and calculated confidence level in a list format. The user can review the displayed proposal information and choose whether or not to adopt it. The evaluation module 216 acquires the proposal results and the user's adoption history as training data and reflects it in the subsequent proposal generation process. Specifically, it dynamically optimizes the entire proposal model by assigning weights to the feature distribution corresponding to proposals with a high adoption rate and decreasing the weights corresponding to proposals with low confidence levels. In this way, by continuously repeating the generation and evaluation of proposals, the action proposal system 1 can adapt to the behavioral characteristics of each user and continuously improve the accuracy and practicality of the proposals.

[0057] In this way, in the proposal output process (S300), the proposal module 215 generates proposal information regarding user behavior based on the importance and evaluation criteria of the tags. The generated proposal information undergoes reliability evaluation using statistical methods by the evaluation module 216, and is output to the user terminal 100 if it meets a predetermined threshold. As a result, the proposal results are output based on probabilistically valid grounds, achieving both objectivity and stability. Furthermore, the evaluation module 216 learnably updates the reliability based on the proposal adoption history and probability distribution structure, dynamically optimizing the entire proposal model. This enables the behavior proposal system 1 to continuously adapt to changes in user behavior while maintaining and improving proposal accuracy over the long term.

[0058] (1.5. Hardware configuration of information processing equipment) Referring to Figure 10, the hardware configuration of an information processing device used as a server 200, etc., will be described. As an example, the information processing device is implemented by the computer 90 shown in Figure 10. The computer 90 may include a CPU 91, ROM 92, RAM 93, storage 94, input interface 95, output interface 96, and communication interface 97.

[0059] The CPU 91 functions as a processor that executes processing. Specifically, the CPU 91 uses the RAM 93 as work memory and the ROM 92 executes programs stored in at least one of the storage 94. During program execution, the CPU 91 controls each component via the system bus 98 and performs various processes. As an example, the CPU 91 functions as a control unit 210.

[0060] ROM92 stores programs that control the operation of computer 90. ROM92 contains the programs necessary for computer 90 to perform each of the processes described above. RAM93 functions as a memory area where the programs stored in ROM92 are loaded.

[0061] The storage 94 stores data necessary for program execution and data obtained through program execution. The storage 94 includes one or more selected from Hard Disk Drives (HDDs) and Solid State Drives (SSDs). As an example, the storage 94 functions as a memory unit 220.

[0062] The input interface (I / F) 95 can connect the computer 90 and the input device 95a. The input interface 95 is, for example, a serial bus interface such as USB. The CPU 91 can read various data from the input device 95a via the input interface 95.

[0063] The output interface (I / F) 96 can connect the computer 90 to the output device 96a. The output interface 96 is a video output interface such as Digital Visual Interface (DVI) or High-Definition Multimedia Interface (HDMI®). The CPU 91 can transmit data to the output device 96a via the output interface 96 and cause the output device 96a to output data.

[0064] The input device 95a is an example of an input means and includes one or more selected from a mouse, keyboard, microphone (voice input), and touchpad. The output device 96a is an example of an output means and includes one or more selected from a display, projector, printer, and speaker. Devices that have the functions of both the input device 95a and the output device 96a, such as a touch panel, may also be used.

[0065] The communication interface (I / F) 97 allows the computer 90 to connect with an external server 97a located outside the computer 90. The communication interface 97 is, for example, a network card such as a LAN card. The CPU 91 can read various data from the external server 97a via the communication interface 97.

[0066] Each process performed by the user terminal 100 or the server 200 may be implemented by a single computer 90, or by the cooperation of multiple computers 90.

[0067] The processing of the various data described above may be recorded as a program that can be executed by a computer on a magnetic disk (flexible disk and hard disk, etc.), an optical disk (CD-ROM, CD-R, CD-RW, DVD-ROM, DVD±R, DVD±RW, etc.), a semiconductor memory, or another non-transitory computer-readable storage medium.

[0068] For example, information recorded on a recording medium can be read by a computer (or embedded system). The recording format (storage format) of the recording medium is arbitrary. For example, a computer reads a program from the recording medium and, based on this program, causes the processor to execute the instructions written in the program. In a computer, program acquisition (or reading) may be performed via a network.

[0069] (1.6.Summary) As described above, the action suggestion system 1 comprises an acquisition module 211, a generation module 212, an assignment module 213, a calculation module 214, a suggestion module 215, and an evaluation module 216. The acquisition module 211 acquires historical information about the user's activities from the user terminal 100, and the generation module 212 divides the historical information into morphological units to generate history segmentation information. The assignment module 213 extracts characteristic words based on the history segmentation information and assigns classification labels (tags). The calculation module 214 calculates the importance of the assigned tags, and the suggestion module 215 outputs suggested actions that the user should take next, based on the classification labels and importance. As a result, the action suggestion system 1 can extract contextually meaningful features from the user's behavior history and generate action suggestions optimized for each individual user.

[0070] Furthermore, the generation module 212 may be configured to store history segmentation information, to which sequential keys have been assigned according to predetermined rules, in the storage unit 220. This ensures that history information is systematically managed while maintaining consistency according to predetermined rules, enabling the generation of suggestions that take into account the context and trends of actions.

[0071] Furthermore, the calculation module 214 may be configured to calculate the probability distribution of classification labels assigned to the history segmentation information as importance and store it in the storage unit 220 along with the order key. This enables probabilistic analysis that includes transition probabilities between behavior labels, rather than just simple frequency analysis, and allows for the quantitative representation of the continuity and relationships of behavioral sequences. As a result, the accuracy of the suggestions is improved, and dynamic suggestion generation that reflects the user's unique behavioral characteristics can be realized.

[0072] Furthermore, the evaluation criteria database 223 may be specified to depend only on the transition patterns along the probability distribution and the order key of the classification labels. This ensures that the proposal generation process by the proposal module 215 is not affected by external factors, and enables objective evaluation based on the statistical characteristics of the historical information itself. In particular, it is possible to suppress the influence of fluctuations in historical data and noise data, and maintain stable proposal performance.

[0073] Furthermore, the evaluation module 216 uses statistical methods to assess the reliability of the proposed information before output. This allows for quantitative verification of the validity and consistency of the proposals before output, making it possible to present only highly reliable proposals to the user. In particular, by applying statistical models such as Bayesian estimation and regression analysis, the reliability of the proposed information can be expressed probabilistically, thereby reducing the number of false proposals.

[0074] Furthermore, the evaluation module 216 may be configured to determine whether the proposed information was adopted and to re-evaluate its reliability based on the result. This allows feedback on the proposal results to be continuously reflected in the learning process, enabling the entire system to evolve in a self-improving manner. In particular, by dynamically adjusting the proposal generation logic based on the proposal adoption rate, it becomes possible to continuously provide highly accurate action suggestions that are adapted to the user's behavioral tendencies.

[0075] Thus, the behavior suggestion system 1 has a configuration that performs a series of operations from morphological analysis of historical information to feature word extraction, tagging, probability distribution calculation, suggestion generation, reliability evaluation, and feedback learning. This improves the accuracy of behavior prediction and suggestions based on the user's past behavior, and continuously improves suggestion performance.

[0076] <2. Other Embodiments> The behavior suggestion system 1 according to this embodiment has been described above, but the application of the technical idea of ​​this disclosure is not limited to the above embodiment. In the above embodiment, an example of making behavior suggestions based on daily report data related to company business was described, but the technical idea of ​​the present invention is not limited thereto and can be applied to any human activity history, such as studying, sports, music, art, and skill acquisition.

[0077] Specifically, the Action Proposal System 1 of this disclosure can be applied in a variety of fields, such as the following: • Analysis of learning progress based on students' learning history and suggestion of next learning content. • Suggestions for optimizing training menus based on athletes' training records. • Presentation of expressive techniques and practice exercises based on the practice histories of musicians and artists. • Support for improving the skills acquisition process in corporate training and technical education. • Task allocation and progress proposals in team activities and collaborative projects.

[0078] For example, when applying the technical concept of this disclosure to the sports field, the acquisition module 211 acquires historical information such as the target athlete's training records, heart rate, distance covered, training intensity, and play data during matches. The generation module 212 divides this data into time units or session units, and the tagging module 213 assigns tags such as "sprint," "pass success rate," "number of shots," and "fatigue level." The calculation module 214 analyzes the frequency of occurrence and fluctuation trends of each tag and calculates importance based on the probability distribution. The suggestion module 215 can then output action suggestions such as "next time, you should take a rest when your heart rate is likely to rise" or "you should focus on passing practice." As a result, athletes can obtain statistically and individually optimized training plans based on their own training data, and coaches can provide strategic guidance while understanding the athlete's condition and tendencies.

[0079] Furthermore, the technical concept of this disclosure is not limited to sports; it can also be applied to performance data in music practice (tempo, mis-touches, dynamic patterns, etc.) or production records in artistic activities (production time, materials used, evaluation scores, etc.). This allows users to objectively analyze their activity history and receive suggestions from the system regarding the next challenges to address and improvement strategies. In other words, the behavior suggestion system of the present invention provides a general-purpose framework for analyzing and evaluating qualitative and quantitative data associated with human activities and suggesting optimal actions.

[0080] Furthermore, in the above embodiment, the server 200 is shown as having various components such as an acquisition module 211, a generation module 212, an assignment module 213, a calculation module 214, a proposal module 215, and an evaluation module 216, and is shown as a configuration that centrally executes each of these processes, but the system is not limited to this. For example, some or all of these processes may be distributed to the user terminal 100.

[0081] Furthermore, instead of using the external AI server 300, a machine learning model built in-house using machine learning may be used, and at least one of the following processes may be executed using this machine learning model: the history segmentation process (S100), the importance calculation process (S200), and the proposal output process (S300). These processes may also be executed using a rule-based approach instead of a machine learning model, or a configuration combining a machine learning model and a rule-based approach may be used. This configuration allows for flexible selection of the optimal estimation and control methods depending on the data characteristics and operating environment.

[0082] While several embodiments of this disclosure have been illustrated above, these embodiments are presented as examples only and are not intended to limit the scope of the invention. These novel embodiments can be implemented in various other forms, and various omissions, substitutions, and modifications can be made without departing from the spirit of the invention. These embodiments and their variations are included in the scope and spirit of the invention, as well as in the claims of the invention and its equivalents. Furthermore, the embodiments described above can be implemented in combination with each other. [Explanation of symbols]

[0083] 1: Proposal Action System, 51: Daily Report Input Screen, 52: Proposal Output Screen, 90: Computer, 91: CPU, 92: ROM, 93: RAM, 94: Storage, 95: Input Interface, 95a: Input Device, 96: Output Interface, 96a: Output Device, 97: Communication Interface, 97a: External Server, 98: System Bus, 100: User Terminal, 200: Server, 210: Control Unit, 211: Acquisition Module, 212: Generation Module, 213: Assignment Module, 214: Calculation Module, 215: Proposal Module, 216: Evaluation Module, 220: Memory Unit, 221: History Segmentation Database, 222: Importance Database, 223: Evaluation Criteria Database, 224: Proposal History Database, 300: External AI Server

Claims

1. A system comprising a control unit and a storage unit, The control unit, A module for acquiring historical information about user activity, A generation module that divides the received historical information into predetermined units and generates historical segmentation information combined with time-series information, A tagging module that extracts characteristic words from the aforementioned history segmentation information and assigns tags as corresponding classification labels, A calculation module that calculates importance based on the probability distribution of the assigned tags, A proposal module that outputs proposal information including the content of the actions indicated by the tags, based on the tags and evaluation criteria which are indicators for evaluating action sequences that are predetermined changes in the time series of actions with respect to the importance of the tags, using a generative AI model to output proposal information including the content of the actions indicated by the tags, which corresponds to action sequences in which the score related to the probability distribution calculated from the changes in the tags satisfies predetermined conditions. A system that includes action suggestions.

2. The action suggestion system according to claim 1, wherein the generation module causes the history division information, to which sequential keys in accordance with predetermined rules have been assigned, to be stored in the storage unit.

3. The aforementioned calculation module is The action suggestion system according to claim 2, wherein the probability distribution of the classification labels assigned to the history segmentation information is calculated as the importance and stored in the storage unit together with the order key.

4. The action suggestion system according to claim 3, wherein the evaluation criteria depend solely on the probability distribution and the mode of transition of the classification labels along the ordinal key.

5. The action suggestion system according to claim 1, wherein the control unit further comprises an evaluation module that evaluates the reliability of the suggested information before output using statistical methods.

6. The action proposal system according to claim 5, wherein the evaluation module determines whether the proposed information has been adopted and evaluates the reliability of the proposed information.

7. An information processing device comprising a control unit and a storage unit, The control unit, A module for acquiring historical information about user activity, A generation module that divides the received historical information into predetermined units and generates historical segmentation information combined with time-series information, A tagging module that extracts characteristic words from the aforementioned history segmentation information and assigns tags as corresponding classification labels, A calculation module that calculates importance based on the probability distribution of the assigned tags, A proposal module that outputs proposal information including the content of the actions indicated by the tags, based on the tags and evaluation criteria which are indicators for evaluating action sequences that are predetermined changes in the time series of actions with respect to the importance of the tags, using a generative AI model to output proposal information including the content of the actions indicated by the tags, which corresponds to action sequences in which the score related to the probability distribution calculated from the changes in the tags satisfies predetermined conditions. Information processing device, including

8. A method for causing an information processing device, which comprises a control unit and a storage unit, to perform processing, The control unit, The acquisition step involves obtaining historical information about the user's activities, A generation step involves dividing the received historical information into predetermined units and generating historical segmented information by combining it with time-series information. The aforementioned history segmentation information is assigned a tag by extracting characteristic words and assigning them as corresponding classification labels, A calculation step of calculating importance based on the probability distribution of the assigned tags, The steps include: outputting suggested information, including the content of the actions indicated by the tags, using a generative AI model based on the tags and evaluation criteria, which are indicators for evaluating action sequences that are predetermined time-series changes in actions relative to the importance of the tags, such that the score related to the probability distribution calculated from the tag changes satisfies predetermined conditions for the action sequences; An information processing method that enables execution.

9. A program that causes an information processing device, which comprises a control unit and a storage unit, to execute processing, The control unit, The acquisition step involves obtaining historical information about the user's activities, A generation step involves dividing the received historical information into predetermined units and generating historical segmented information by combining it with time-series information. The aforementioned history segmentation information is assigned a tag by extracting characteristic words and assigning them as corresponding classification labels, A calculation step of calculating importance based on the probability distribution of the assigned tags, The steps include: outputting suggested information, including the content of the actions indicated by the tags, using a generative AI model based on the tags and evaluation criteria, which are indicators for evaluating action sequences that are predetermined time-series changes in actions relative to the importance of the tags, such that the score related to the probability distribution calculated from the tag changes satisfies predetermined conditions for the action sequences; A program that executes something.

Images