Information processing device, information processing method, and computer program

Abstract

An information processing device that dynamically reconfigures an information structure according to history and external information is provided. [Solution] The information processing device includes a score evaluation unit that calculates link scores between phrase nodes, a structure reconstruction unit that reconstructs a presentation structure based on the link scores, and a storage unit that stores the reconstructed phrase network structure. The score evaluation unit calculates a history score based on at least one of co-occurrence frequency, temporal proximity, and selection order. The structure reconstruction means deletes the corresponding link when the calculated link score falls below a predetermined threshold T, and reconnects the corresponding link when the link score of the link again exceeds threshold T.

Description

[Technical Field]

[0001] The present invention relates to an information processing device, an information processing method, and a computer program. [Background technology]

[0002] As shown in Reference 1, technology related to language models trained from corpora is known. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 7626555

[0004] For example, in a phrase recommendation system, a link score is typically evaluated based on the frequency of co-occurrence between phrase nodes and their temporal proximity, and the link is strengthened if the score exceeds a certain threshold. For example, if a phrase is frequently selected immediately after another phrase, a process is performed to strengthen the link strength between the two phrase nodes. Another known technique is to apply time decay correction to the co-occurrence frequency to reduce the influence of older history. However, conventional methods such as those described above cannot distinguish between small and large fluctuations in the link score, resulting in delays in structural reconstruction even when user interests change significantly, leaving room for improvement. Summary of the Invention [Problem to be solved by the invention]

[0005] An object of the present invention is to provide an information processing device that dynamically reconfigures an information structure in accordance with history and external information. [Means for solving the problem]

[0006] The information processing device includes a score evaluation unit that calculates link scores between phrase nodes, a structure reconstruction unit that reconstructs a presentation structure based on the link scores, and a storage unit that stores the reconstructed phrase network structure. The score evaluation unit calculates a history score based on at least one of co-occurrence frequency, temporal proximity, and selection order. The structure reconstruction means deletes the corresponding link when the calculated link score falls below a predetermined threshold T, and reconnects the corresponding link when the link score of the link again exceeds threshold T. [Brief explanation of the drawings]

[0007] [Figure 1] FIG. 1 illustrates an example of a configuration of an information processing device. [Figure 2] FIG. 10 is a diagram illustrating an example of a processing flow of an information processing device. [Figure 3] FIG. 1 illustrates an example of a hardware configuration of an information processing device. [Figure 4] FIG. 1 is a diagram illustrating an example of a network structure related to earthquakes. [Figure 5] FIG. 10 is a diagram showing an example of a change in the presentation structure. [Figure 6] FIG. 1 is a diagram showing an example of a network structure related to oshikatsu. [Figure 7] FIG. 10 is a diagram illustrating an example of a process for reevaluating a deleted link. [Figure 8] FIG. 10 is a diagram illustrating an example of a reconstruction process using ΔS. [Figure 9] FIG. 10 is a diagram illustrating an example of a fixed link process. DETAILED DESCRIPTION OF THE INVENTION

[0008] Hereinafter, an embodiment of the present invention will be described with reference to the drawings. The definitions of terms used in this specification are as follows: "Phrase" refers to a linguistic expression selected or input by a user (e.g., "earthquake," "tsunami," "earthquake disaster special law," etc.), and includes data units that are structured as phrase nodes on an information processing device. A "phrase node" refers to a point corresponding to a certain "phrase" in the structure database 150. Phrase nodes are connected to each other by "links (lines)" based on a predetermined relationship. Unless otherwise specified, "score" refers to the "link score" that evaluates the connection relationship between phrase nodes. Examples of this are the "history score" and the "corrected score." "Historical Score" means a link score calculated based on historical information. "Adjusted score" refers to a weighted evaluation value used to supplement the historical score based on external information. "Phrase network" refers to a network structure formed by the connections between phrase nodes. "Reconstruction" refers to the process of restructuring the link structure between phrase nodes or the presentation structure, and unless otherwise specified, means "structural reconstruction."

[0009] (1. Comparison of the Prior Art and the Present Invention) Conventional phrase recommendation systems have the problem that they cannot update their presentation structure in response to changes in the user's operation history or the external environment, and the presented information cannot respond to the user's interests or environment, making them impractical.

[0010] Specifically, the conventional method has at least the following three technical limitations. (i) It is not possible to distinguish between small and large changes in link scores, which delays the restructuring process even when user interests change significantly. (ii) There is no mechanism for adjusting scores by simultaneously taking into account the importance of external events and the reliability of the information source, so appropriate words cannot be prioritized in emergencies. (iii) There is no mechanism to restore deleted links through history learning, so useful related words are lost forever due to a temporary loss of interest.

[0011] As such, conventional recommendation methods that evaluate link scores between phrase nodes based on co-occurrence frequency and temporal proximity have difficulty appropriately reflecting external information such as user attributes and usage environment, and do not have a mechanism for reusing links that have been deleted due to temporary changes in phrase scores.In addition, there is no process for dynamically updating the presentation structure in response to fluctuations in the scores between phrases, which causes a problem in that the phrase network structure is slow to follow changes in interest.

[0012] The present invention employs a configuration in which the connection structure between suggested phrase nodes itself is changed based on the user's operation history. Specifically, through a score evaluation process for inter-node links, weak links are deleted and strong links are emphasized or newly generated, thereby realizing a structurally adaptable phrase network.

[0013] Specifically, in order to solve the above problems (i) to (iii), the present invention provides: (A) A conditional logic that monitors the link score fluctuation amount ΔS, immediately reconfigures when ΔS > ΔT, and fixes the link when ΔS ≦ ΔT continues. (B) A method for synthesizing a corrected score based on the "urgency x reliability" assigned to external information. (C) A log learning method that keeps a score history of deleted links and autonomously reconnects them when the threshold is exceeded again. To provide an information processing device that integrates the above.

[0014] (2. Embodiment) FIG. 1 is a diagram showing the configuration of an information processing device 100 according to the present invention, and FIG. 2 shows the processing flow thereof.

[0015] The information processing device 100 includes an input interface 110 , a history acquisition unit 120 , a score evaluation unit 130 , a structure reconstruction unit 140 , a structure database 150 , and an output interface 160 .

[0016] The input interface 110 has a function of receiving a user's selected phrase or search request, and accepts input from the user of a phrase or keyword (hereinafter referred to as a target phrase) for which information is to be presented.

[0017] The history acquisition unit 120 acquires history information based on the selection history of target words acquired via the input interface 110 or the past presentation history recorded in the structure database 150. The acquired history information includes chronological attributes such as selected words, selection times, and selection order. In this specification, "history information" refers to data acquired regarding the selection history described above, and is used as input information for score evaluation processing, etc.

[0018] The score evaluation unit 130 calculates a score for each link between phrase nodes based on the history information acquired by the history acquisition unit 120, taking into consideration evaluation indices such as the co-occurrence frequency of target phrases, temporal proximity, and selection order. The score evaluation unit 130 also acquires multiple evaluation indices such as history scores and corrected scores, and calculates or weighted-combines evaluation values ​​for each link between phrase nodes. In this way, the score evaluation unit 130 performs all calculation, composition, and evaluation processes related to scores.

[0019] As an example, the score evaluation unit 130 receives external information (disaster information, time period, terminal type, environmental sensor information, etc.) as input and calculates a corrected score for the history score. In this specification, "external information" refers to information related to environmental or system factors that is acquired independently of the user's operation history, and includes disaster occurrence status, weather, time, location information (GPS, etc.), terminal type, sensor data (acceleration, temperature, etc.), etc.

[0020] The structure reconfiguration unit 140 performs a link structure reconfiguration process based on the scores calculated by the score evaluation unit 130. Links with scores below a predetermined threshold are deleted, and links with scores equal to or greater than the threshold are maintained or strengthened. New links are also generated as necessary.

[0021] The reconstructed network structure is stored in the structure database 150 and becomes the target of the next score evaluation process. The reconstructed result is written back to the structure database and referenced as input for the next score evaluation, thereby forming a sequential update loop of the presentation function as shown in Figure 1.

[0022] The output interface 160 presents a group of related words to the user based on the reconstructed network structure. The presented group of words is sorted in descending order of visibility and relevance, and guides the user's next selection behavior.

[0023] With the above configuration, the information processing device 100 constructs a network structure adapted to history and external information, and presents a group of words whose structure has been dynamically and autonomously optimized.

[0024] In the processing flow shown in Fig. 2, in step S101, a word input or selection from the user is received via input interface 110. Examples of this include an operation in which the user inputs a word into a search window displayed on a display serving as input interface 110, or an operation in which the user selects a word from among previously presented words.

[0025] In step S102, the history acquisition unit 120 acquires selected words and phrases, their selection times and rankings, etc. in chronological order based on the input word and phrase information, and also acquires related past selection histories from the structure database 150.

[0026] In step S103, the score evaluation unit 130 calculates the score between each node based on the sequence of selected words and phrases acquired by the history acquisition unit 120. The score is evaluated by weighting factors such as the frequency of co-occurrence between words, the selection interval, and the proximity of the selection order. As an example, the history score R can be calculated using the following (Equation 1):

[0027] R = w1× f_co + w2× f_time + w3× f_order (Equation 1)

[0028] In (Equation 1), f_co is a co-occurrence frequency score, normalized based on the number of times that terms A and B appear together within the same session (a predetermined unit of user search behavior). f_time is the choice time proximity score, modeled with a decay function such as exp(-ΔT) based on the difference in choice time. f_order is the selection order agreement score, which represents the probability that the order of phrases A → B was maintained. Furthermore, w1, w2, and w3 are weights for each element, and satisfy the relationship w1+ w2+ w3=1.

[0029] Here, f_co is a normalized score of the co-occurrence frequency between words (e.g., 0.0 to 1.0), f_time is the inverse of the selection timestamp or a decay score based on the passage of time (e.g., exp(-λt)), and f_order is calculated as the inverse score according to the selection order (e.g., 1 / order).

[0030] For example, if w1=0.5, w2=0.3, w3=0.2, f_co=0.8, f_time=0.5, and f_order=1.0, the following formula (Formula 2) is obtained.

[0031] R = 0.5×0.8 + 0.3×0.5 + 0.2×1.0 = 0.4 + 0.15 + 0.2 = 0.75 (Equation 2)

[0032] Furthermore, the history score R can be configured as follows: For example, when w1=0.4, w2=0.4, w3=0.2, f_co=0.5, f_time=0.9, and f_order=0.3, the following (Equation 3) is obtained.

[0033] R = 0.4×0.5 + 0.4×0.9 + 0.2×0.3 = 0.20 + 0.36 + 0.06 = 0.62 (Equation 3)

[0034] This makes it easier for the link scores of phrases with high temporal proximity to increase, strengthening the connections within the network.

[0035] In this way, multiple behavioral indicators are integrated into a score and applied to the reconstruction. These coefficients can be manually set depending on the operational purpose of the system, allowing the evaluation policy to be controlled according to the application.

[0036] Furthermore, in the present invention, a process may be performed to dynamically optimize these weighting coefficients themselves in response to changes in the user's history and selection tendency. For example, the coefficients themselves may be reevaluated as optimization targets, such as adjusting w1 to a higher value when the co-occurrence frequency is steadily increasing, or decreasing the weighting of w3 when the effectiveness of the selection order is decreasing. This dynamic control allows the history score to be constantly evaluated in accordance with the user's behavioral tendency. The initial values ​​of w1 to w3 may be set, for example, as w1 = 0.4, w2 = 0.3, and w3 = 0.3. Each time the user's selection history is accumulated, the proportion of each coefficient may be adjusted based on the success rate of search results and order stability in the most recent 10 sessions.

[0037] In step S103, if external information is available, a corrected score is calculated according to the information (e.g., disaster alerts, time zones, location information, etc.). The score evaluation unit 130 integrates the history score and the corrected score, and executes a process of calculating a final score.

[0038] The means for acquiring external information may involve acquiring data via HTTP communication or RESTful API from external servers such as disaster information providers, the Japan Meteorological Agency API, location information services, etc. Sensor information (such as acceleration, position, and temperature) installed in the device may also be configured to be acquired via the system APIs of various OSs, such as Android's SensorManager or iOS's CoreMotion.

[0039] In the score calculation process, the score evaluation unit 130 weights the history score and the corrected score. The weighting coefficient is dynamically adjusted depending on the reliability and urgency of the external information. For example, the corrected score C is calculated as C = 0.8 × 1.2 = 0.96 using the reliability (confidence = 0.8) and urgency (urgency = 1.2) included in the response header from the disaster information API.

[0040] In step S104, the structure reconfiguration unit 140 performs a link structure reconfiguration process based on the score evaluation result. Links with scores below a predetermined threshold are deleted, and links with scores equal to or greater than the threshold are retained or newly generated as necessary. The reconfiguration process branches depending on the score; if the history score exceeds the threshold, a normal reconfiguration is applied, and if it falls below the threshold, a supplemental reconfiguration that takes the correction score into account is selected.

[0041] Specifically, the reconstruction process branches into two paths depending on the score evaluation result. If the score is equal to or greater than a predetermined threshold, "S104(a): normal reconstruction" is executed. In normal reconstruction, a process is executed to maintain links with high scores based on the history score and delete links with low scores. On the other hand, if the score is less than the threshold, "S104(b): reinforced reconstruction" is executed, and a correction score is applied in addition to the history score. In reinforced reconstruction, links related to the external environment (e.g., disaster information, time of day, etc.) are dynamically reinforced to build a consistent presentation structure even in situations where there is insufficient history.

[0042] In this case, the structure reconfiguration unit 140 may dynamically adjust the threshold T according to the overall score distribution and the history accumulation status, rather than setting it as a fixed value. For example, the structure reconfiguration unit 140 may perform reconfiguration based on data statistics, such as using the median or variance of the link scores as indicators and deleting links that fall in the bottom 30%.

[0043] Furthermore, if the user's selection tendency is consistently stable, the structure reconstructor 140 may adjust the threshold to prioritize structural stability, while if a sudden change in interest is observed, the structure reconstructor 140 may autonomously adjust the threshold to increase the aggressiveness of the reconstruction.

[0044] The reconstructed network structure is sent to the output interface 160 as a presented structure and presented to the user. Note that even if the structure is not reconstructed, the presented structure is determined based on the internal state, such as determining the stability of the links and updating the history of fluctuations in the link score fluctuation amount ΔS.

[0045] Furthermore, as shown by the feedback line in Figure 2, the latest presented structure is immediately recorded in the structure database 150 and reacquired by the history acquisition unit 120 through the user's selection behavior, and is usually reflected in the score evaluation process within N presentation periods, forming a feedback loop for structure update.

[0046] 3 is a diagram showing an example of the hardware configuration of the information processing device 100. Components include a processor 210, a memory 220, a storage 230, a communication interface 240, an input device 250, and an output device 260, and the functions of each unit are implemented as software modules.

[0047] The processor 210 comprehensively controls the functions of the history acquisition unit 120, score evaluation unit 130, and structure reconstruction unit 140, and the memory 220 stores the structure database 150 and history information. This enables continuous reconstruction processing. Each component (history acquisition unit, score evaluation unit, structure reconstruction unit) is implemented as a software module executed on the processor 210 in the information processing device 100. Furthermore, these functions are stored as program code stored in the memory 220 or storage 230, and the program code is loaded onto the memory 220 and executed at startup. The structure database 150 may be stored in the memory 220 or storage 230, or may be implemented as a separate module or an external device as needed.

[0048] The communication interface 240 may be configured to receive disaster information, weather data, etc. from an external server, and the acquired information is used to calculate the corrected score. The input device 250 and the output device 260 function as the input interface 110 and the output interface 160.

[0049] A specific example of processing by the structure reconstructing unit 140 in this embodiment will be described below. Figure 4 shows an example of a phrase network structure between candidate phrase nodes. Node A (earthquake) is at the center, and other phrase nodes are connected, such as node B (disaster prevention), node C (seismic intensity), node D (tsunami), node E (emergency evacuation bag), and node F (earthquake disaster special law).

[0050] In this initial structure, links between phrase nodes are generated based on their past co-occurrences, and each link is assigned a historical score. The thickness of the link reflects the score, with thick links indicating high scores and thin links indicating low scores.

[0051] Based on the link scores calculated by the score evaluation unit 130, the structure reconstructing unit 140 targets links with low scores (for example, between A and E, between C and F) for deletion, and maintains links with high scores (between A and D, between A and B, etc.).

[0052] This link selection process allows the phrase network to converge to a form that is consistent with the user's operational tendencies. Since phrase nodes in the past history are not deleted and only links are selected, loss of information is avoided.

[0053] FIG. 4(b) shows an example of a phrase network structure after the structure reconstructor 140 deletes low-score links from the initial structure shown in FIG. 4(a) based on the link score calculation results by the score evaluation unit 130. Specifically, the links between node A "earthquake" and node E "emergency evacuation bag," and between node C "seismic intensity" and node F "earthquake disaster special law," are determined to have scores below a predetermined threshold and are therefore subject to deletion. This removes connections between phrase nodes with low semantic relevance in the user's selection history, and the reconstruction proceeds to avoid redundant presentation structures. In this embodiment, when T = 0.3 is used as an example of the threshold T, connections between nodes with link scores below this threshold are subject to deletion. This threshold T is applied in combination with each link score calculated based on the score evaluation formula shown in (Equation 1). The circled symbols 1 and 2 in FIG. 4(b) indicate the links between nodes A to E and between nodes C to F, respectively, which are subject to the link deletion process.

[0054] For example, in Figure 4(b), node A "earthquake" and node F "earthquake disaster special law" are not directly connected, but if the co-occurrence frequency or order consistency of the two terms increases in the user's selection history, a new link will be generated as shown in Figure 4(c).

[0055] Similarly, for the link between node D "Tsunami" and node F "Earthquake Disaster Special Measures Act" in Figure 4(b), if the history score increases, it is strengthened as shown in Figure 4(c), and this is visually reflected as the thickness of the link.

[0056] FIG. 5 shows the process of changing the presentation structure in the present invention, with FIG. 5(a) showing the initial structure, FIG. 5(b) showing the stage where score evaluation is applied after recording the selection history, and FIG. 5(c) showing the presentation structure after undergoing reconstruction processing. When the user's selection history is recorded, the score evaluation unit 130 of the information processing device 100 performs score evaluation based on the history, and the link structure between word nodes is reconstructed according to the results. The presentation structure after reconstruction is configured as a network in which the connection relationships have changed according to the link scores, and a word arrangement such as that shown in FIG. 5(c) is formed.

[0057] For example, if the user shows a strong interest in node F "Earthquake Disaster Special Measures Act," the structure reconfiguration unit 140 of the information processing device 100 relocates node A "Earthquake," which was placed in the center of the initial structure, to a position below node F "Earthquake Disaster Special Measures Act." Then, by repeating the dynamic link selection process, a presentation structure is formed according to the user's selection tendencies.

[0058] Even if a link is deleted in the reconstruction, its evaluation history is maintained in the structure database 150, and when the user's interest later turns again to the node, a new link is generated based on the evaluation history.

[0059] For example, if "emergency bags" start to attract users' attention through news reports or disaster prevention campaigns, a new selection history for that node will be generated, resulting in an increase in the score and reconnection of deleted links.

[0060] The structure reconstructing unit 140 of the information processing device 100 historically records link structure information before and after reconstruction in the structure database 150, and cooperates with a function to visualize and analyze the historical information as time-series network changes. Furthermore, when a user performs a word selection operation a predetermined number of times (e.g., 10 times) or more, the reconstructing process may be automatically initiated. Furthermore, when the accumulation of selection operations reaches within a predetermined time interval (e.g., 30 minutes), the reconstructing process may be executed in a concentrated manner, thereby functioning as a time-linked reconstructing condition.

[0061] Furthermore, if the same link structure is reconstructed a certain number of times (e.g., three times) or more as a result of the reconstruction process, the link can be identified as a "non-changing link" and designated as a "fixed link" that will not be subject to deletion or reconnection in subsequent reconstruction processes. In this way, for links whose link score fluctuation amount ΔS remains stable and is below a threshold, the link can be treated as a fixed link and excluded from evaluation in subsequent reconstruction processes. As a result, the overall processing load is reduced and the stability of the word candidate presentation to the user is improved.

[0062] A link is determined to be a "fixed link" if its link score fluctuation amount ΔS (Delta S) is equal to or less than a predetermined fluctuation threshold ΔT (Delta T) and if that state continues for a certain number of times (e.g., three or more times). Fixed links are stored as flag information in the structure database 150 and are excluded from re-evaluation targets in subsequent reconstruction processes. ΔS is usually defined in the range of 0.0 to 1.0, and ΔT is set in the range of 0.1 to 0.9, for example. Furthermore, if the state of ΔS<ΔT continues multiple times, for example, three or more times, the link is determined to have a stable structure and is considered a fixed link.

[0063] When the load of the reconstruction process is high, the structure reconstruction unit 140 of the information processing device 100 may perform only the score evaluation first and temporarily store the result in a buffer. Then, just before presentation, the reconstruction process may be performed all at once based on the stored score results.

[0064] Furthermore, the score evaluation unit 130 of the information processing device 100 dynamically adjusts the combination ratio of the history score and the corrected score depending on the reliability of the external information, the time period, and the user attributes.

[0065] For example, the score evaluation unit 130 of the information processing device 100 dynamically controls the balance of structure presentation, such as by increasing the weight of the corrected score in an emergency such as a disaster alert and by focusing on the history score in normal times.

[0066] The score evaluation unit 130 may normally use a method of applying a weighted average to the history score and the corrected score, but may also combine scores by combining conditional branching selection control that invalidates one of the scores under certain conditions.

[0067] Even in a situation where the correction score is not applied or where external information cannot be acquired, the structure reconstructing unit 140 may continue the reconstruction process based on the history score.

[0068] The score evaluation unit 130 may integrate the link structure based on the history score and the structure control by the corrected score based on external information, and reflect the results in the score synthesis process.

[0069] The structure reconstructing unit 140 may also adjust the arrangement order of the word nodes and the priority of the related nodes based on the scores, and reflect the results in the arrangement order of the presentation structure. Furthermore, the presentation order of the word nodes may be adjusted according to the centrality in the network or the hierarchical structure.

[0070] For example, if a user repeatedly selects the word "earthquake," and transitions to words such as "tsunami" and "disaster prevention" are observed continuously, the link score between node A "earthquake" and node D "tsunami" or node B "disaster prevention" will increase. On the other hand, if the co-occurrence frequency with "emergency evacuation bag" and "earthquake disaster special law" is low, the links to these will be subject to deletion and separated from the presentation structure.

[0071] Such changes in the link structure are automatically made by the collaborative processing of the score evaluation unit 130 and the structure reconstruction unit 140, and the network structure stored in the structure database 150 is updated.

[0072] If the score exceeds the threshold again, the structure reconstructing unit 140 reconnects the previously deleted link. For example, if the phrase "Earthquake Disaster Special Law" temporarily attracts attention due to news reports about "earthquakes" and starts to appear in the user's selection history, the structure reconstructing unit 140 reevaluates the score of the link and reconnects it to the presentation structure.

[0073] The structure reconfiguration unit 140 may also switch between executing the link reconfiguration process in real time and executing it collectively at regular intervals depending on the conditions.

[0074] First, when the change in the score evaluation exceeds a certain threshold, the structure reconfiguration unit 140 executes the link reconfiguration process in real time. For example, this corresponds to the case where the score fluctuation amount ΔS = S_current - S_prev exceeds a predetermined threshold T.

[0075] Meanwhile, the fluctuation threshold ΔT used in the condition linkage of the reconstruction process is updated periodically based on score fluctuations. For example, the statistical distribution (mean, standard deviation, interquartile range, etc.) of the score fluctuation amount ΔS is evaluated for the most recent fluctuation history of the link score, and ΔT is adaptively corrected according to the trend of score fluctuations over a certain period of time in the past, thereby performing reconstruction according to the stability of score changes. ΔT is calculated, for example, using a distribution index of ΔS accumulated over a certain period of time in the past, and may typically be calculated using the interquartile range (IQR). Specifically, fluctuations outside the range of ±IQR / 2 of the median of ΔS may be considered "abnormal fluctuations," and this threshold may be set as ΔT to dynamically adjust the trigger criteria for structural reconstruction.

[0076] Alternatively, ΔT may be calculated using Kullback-Leibler divergence (KL divergence), which quantifies the information-quantity deviation between the historical distribution of ΔS and the ideal distribution. By using such specifications, it is possible to apply this to adaptive control that proactively triggers reconfiguration when the distribution of anomaly scores changes significantly.

[0077] Furthermore, the information processing device 100 may manage the history information recorded in the structure database 150 individually for each user, and may further integrate the history information obtained from multiple users and manage it centrally as a user group history.

[0078] In this configuration using user group history, the score evaluation unit 130 weights and combines the individual user history score and the user group history score to generate a link score that reflects both individuality and overall trends.

[0079] In the process of correcting the score, the score evaluation unit 130 sets application conditions and control policies for specific score items and switches the adjustment of values ​​as necessary. These processes are performed using components common to this embodiment.

[0080] The external information includes disaster information (earthquake, tsunami, etc.), weather information, time information, location information, terminal type information, etc. This information is used by the score evaluation unit 130 in the calculation process of the corrected score.

[0081] External information may be obtained from an external information server via internet communication (e.g. HTTPS, Web API, etc.). Sensors (acceleration, position, temperature, etc.) within the user device may also be obtained using a local API.

[0082] The score evaluation unit 130 applies a correction value to the score between phrase nodes based on the external information. The correction is performed on a link-by-link basis, and in the event of a disaster, the link scores of disaster-related phrases are preferentially increased.

[0083] For example, when an earthquake occurs, correction scores are added between phrase nodes such as "seismic intensity," "tsunami," and "disaster prevention" and the central node "earthquake," and the score of the link increases when reconstructed.

[0084] The corrected score is combined with the history score and used as the final link score in the reconstruction process. The combination uses a weighted average, and the score evaluation unit 130 dynamically adjusts the weight for each score depending on the reliability and urgency of the external information.

[0085] In this case, the final score S may be calculated using the history score R and the corrected score C by the following formula:

[0086] S = α × R + β × C (where α + β = 1) (Equation 4)

[0087] As an example of setting the weighting coefficients α and β, under normal circumstances, α = 0.8 and β = 0.2, and phrases are presented with a focus on the history score R. On the other hand, in situations of high urgency such as when a disaster occurs, settings are applied that prioritize the correction score C, such as β = 0.7 to 0.9 and α = 0.1 to 0.3. For example, if the normal configuration is history score R = 0.75, correction score S_ext = 0.96, and synthesis coefficient α = 0.8, then the following (Equation 5) is obtained.

[0088] S_final = 0.8×0.75 + 0.2×0.96 = 0.792 (Equation 5)

[0089] On the other hand, in the event of a disaster, if the history score R = 0.62, the correction score S_ext = 0.99, and α = 0.3 (β = 0.7) are set, the following (Equation 6) is obtained, and reconstruction is performed based on the correction score.

[0090] S_final = 0.3×0.62 + 0.7×0.99 = 0.879 (Equation 6)

[0091] The information processing device 100 may also perform clustering based on user selection trends and history characteristics, and apply different α or ΔT values ​​to each user cluster. The control balance of the presentation structure is automatically adjusted according to the sensitivity of score fluctuations and the tendency of history dependency. This clustering process is periodically performed by the score evaluation unit 130, and the control parameters are optimized according to the attributes of the user group. The clustering process uses, for example, the k-means method or DBSCAN method, and updates the threshold based on the distribution of score fluctuation amount ΔS.

[0092] Furthermore, a correction score may be calculated based on the reliability and urgency assigned to each external information source, and this may be weighted and combined with the history score to obtain a final score, which may then be used to control the presentation structure. For example, for a disaster information source such as an Earthquake Early Warning System (EWS), the reliability weight_r is set to 0.8, and the urgency weight_u is set to 1.2, and the correction score S_ext is calculated using the following formula (7):

[0093] S_ext = weight_r × weight_u = 0.8 × 1.2 = 0.96 (Equation 7)

[0094] In the present invention, these weighting coefficients are not fixed values ​​either, and may be automatically adjusted by the score evaluation unit 130 in the information processing device 100 according to the user's operation status and the reliability and urgency of the external information.

[0095] For example, when a user is new and has insufficient history, the weighting factor β of the correction score C may be increased to strengthen judgments based on external information. On the other hand, when a user has accumulated a wealth of history and the consistency of the context can be clearly ensured, the weighting factor α of the history score R may be increased, shifting to a stable presentation structure driven by history.

[0096] Furthermore, under temporary circumstances such as a disaster or abnormal weather, the behavior may be switched by controlling the score composition ratio, such as by preferentially reflecting the corrected score and then using the history score as the driving force in stable situations. By adopting such specifications, even when external information cannot be obtained, the reconstruction process based only on the history score continues, maintaining the stability of the presentation structure.

[0097] On the other hand, if correction based on external information is overemphasized, excessive fluctuations may occur in the presented structure, so upper limits may be set on the range of change in the correction score and the frequency of reconstruction to suppress structural disturbances.

[0098] Furthermore, the target words and strength of the corrected score may be dynamically changed depending on the user attributes (age group, occupation type, etc.).

[0099] Below is simple pseudocode showing the calculation of the corrected score S_ext and the process of combining it with the historical score. Weighted combination of historical and corrected scores def compute_final_score(R_hist, weight_r, weight_u, alpha=0.6): S_ext = weight_r * weight_u (calculation of the corrected score) S_final = alpha * R_hist + (1 - alpha) * S_ext return S_final

[0100] Another example of a network structure will be described with reference to FIG. When users' interests are concentrated in a particular cultural or hobby area, a genre-specific phrase network may be formed. Figure 6(a) shows an example of a phrase network structure for "oshikatsu" (supporting someone)-related phrases, which are popular among young users. Node A "oshikatsu" is at the center, and phrase nodes such as Node B "godly response," Node C "best," Node D "oshimen" (supporting member), and Node E "swamp" are connected.

[0101] The structure reconstruction unit 140 calculates a link score based on the frequency of co-occurrence and temporal proximity between these words in the selection history, and evaluates the strength of the link between node A "oshikatsu" and other nodes. For example, if node A and node D "oshimen" co-occur frequently and stably, the link score increases and is configured as a high-score link represented by a thick line. On the other hand, if the co-occurrence between node A "oshikatsu" and node E "swamp" is temporary, the link score is low and it is maintained as a low-score link represented by a thin line.

[0102] In a phrase network that contains many such emotional words and empathetic expressions, link scores tend to fluctuate significantly in a short period of time due to temporary changes in the user's behavior. The structure reconstructor 140 performs real-time evaluation of such score fluctuations and updates the display order and connection priority of phrase nodes in the presentation structure.

[0103] Furthermore, for a word or phrase that has been repeatedly observed to have a certain co-occurrence pattern, the structure reconstructing unit 140 automatically generates the word or phrase as a new word or phrase node and forms a new link between the word or phrase and an existing node.

[0104] 6, the structure reconstructing unit 140 automatically generates a new phrase node based on the accumulation of certain co-occurrence patterns and selection frequencies. For example, for an unregistered phrase G, a phrase node is generated if all of the following conditions are met: (1) The number of co-occurring words in the selection history is equal to or greater than a threshold (e.g., three words). (2) The number of selections within a certain period (e.g., 24 hours) is greater than or equal to a threshold (e.g., 10 times). (3) At least one word in the co-occurring phrase is connected to a high-scoring node in the network.

[0105] With this configuration, even if a new word such as "treasury" begins to be used in a short period of time, the structure reconstructing unit 140 will autonomously expand the word network structure and keep up with the latest vocabulary trends.

[0106] As shown in Figure 6(b), due to the accumulation of history scores and changes in co-occurrence patterns, the structure reconstruction unit 140 generates node F "precious" and forms links between this node and "oshimen" and "saikou." At this time, because "precious" has come to be selected as a word that expresses a stronger emotion than "saikou," the link score between node C "saikou" and node A "oshikatsu" decreases relatively and is represented by a thin line, while the link score of node F "precious" increases, resulting in a high-score link represented by a thick line.

[0107] The flow of the history reevaluation and reconnection process for a deleted link will be described with reference to FIG. If a user's interest in a certain word temporarily wanes, the link score for that word will decrease. The information processing device 100 detects this score change and deletes the link. History information about the deleted link is stored in the structure database 150. For example, even if the link between the word node "vocabulary" and the node "favorite member" was deleted in the past, if co-occurrence selection is repeated again due to a change in the user's interest and the history score exceeds the threshold, the link will be reconnected and the presentation structure will be updated.

[0108] An example of the reconstruction process based on the link score variation amount ΔS will be described with reference to FIG. The information processing device 100 also records the link score fluctuation amount ΔS between phrase nodes, and executes a reconstruction process when ΔS exceeds a predetermined difference threshold ΔT. For example, if a sudden fluctuation in ΔS is detected in the score relationship with the node “highest” immediately after the appearance of the phrase node “precious,” the reconstruction process is immediately initiated, and the link configuration within the presentation structure is changed.

[0109] When a deleted link is reconnected, the information processing device 100 performs log control to record the number of reconfigurations for that link and the history of changes in the link score. This log includes the time series transition of the ΔS value for each link, the number of reconnections that occurred, the time point when the fixed link was designated, and the transition of the evaluation score R. Furthermore, based on the log, the stability of each link is evaluated, and a link whose ΔS has been below ΔT for a predetermined number of times is determined to be a "no-change link" and designated as a fixed link.

[0110] An example of the fixed link process will be described with reference to FIG. For example, after accumulating reconfiguration logs for 200 links, the information processing device 100 reconnects 78 links, and those in which the ΔS<ΔT state continues for three or more times are designated as fixed links and recorded. In this way, the relationship between the number of links excluded from the reconfiguration process and the total number of reconfigurations is dynamically recorded. The log of ΔS variation values ​​for each link is configured to be managed in chronological order and is saved, for example, in a fixed-size ring buffer format or a time-ordered table structure. Each ΔS value is assigned a timestamp, and after a certain period or after a certain number of logs have been recorded, the oldest log is overwritten and used to determine stability.

[0111] (3. Conclusion) According to the present invention, link structures between phrase nodes are dynamically reconfigured based on user selection history and external information, enabling flexible phrase presentation according to usage conditions and emotional trends while maintaining contextual consistency. Furthermore, a weighted combination of history scores and correction scores enables a responsive presentation structure to be applied even when history is not accumulated, ensuring presentation consistency and contextual stability. Furthermore, the reconfiguration process is controlled in real time or in a batch manner, concentrating processing immediately before presentation reduces computational load and improves the accuracy of structure updates. Conditional branching control based on score fluctuations and link selection based on history logs simultaneously adapts to changes in interest and prevents contextual inconsistencies. In particular, links with stable, highly rated scores are designated as fixed links and excluded from future reconfiguration, reducing the frequency of reconfiguration and improving processing efficiency. The score evaluation and structure reconfiguration configurations used in the present invention not only optimize the presentation structure in a phrase network, but also provide a general-purpose technical platform for dynamically controlling information structures through the management of score fluctuations and stable structures. Therefore, this configuration has the potential to be applied in the future to the reconstruction of entangled structures and the control of transmission efficiency in information transmission systems such as quantum communication networks, and is positioned as one of the fundamental technologies for information structure control in general.

[0112] The configuration of the present invention can be applied to various information presentation systems to reconstruct the network structure based on the history score, the corrected score, and the score fluctuation amount ΔS. An example of its operational application is shown below. For example, in a disaster information application, correction scores based on disaster alerts and weather changes are applied to dynamically strengthen the link structure of phrases such as "tsunami," "evacuation," and "emergency evacuation bag." If user interests suddenly change, the phrase network is instantly reconstructed according to ΔS. In the recommendation system of an e-commerce site, the link score of products that users have previously viewed but not purchased is temporarily reduced and the products are removed from the presentation structure. On the other hand, if the correction score increases due to external information such as the start of a sale or restocking, the link is reconnected and the re-presentation process based on ΔS is activated. In detecting trending terms on SNS, the link score is dynamically updated based on the term selection history obtained from user posts, and if the co-occurrence pattern with a newly emerged term exceeds a predetermined threshold, a reconfiguration is triggered based on ΔS, and a new term node is added to the network. In the call center FAQ construction system, scores between query terms are accumulated based on the selection history, and reconstruction processing is concentrated on links with large ΔS, thereby achieving both adaptability and responsiveness in the FAQ structure. In addition, the priority order of options is updated according to the chronological history. In prompt generation support for large-scale language models (LLMs), a synthesis process is performed on word scores based on the prompt design history and correction scores based on the external system state (dialogue history and objective conditions), and the optimal reconfiguration timing is dynamically controlled by ΔS. In an environmentally responsive information presentation system, the correction score fluctuates based on dynamic changes in the external environment, such as disaster information and sensor information, and the word link structure is reconfigured according to changes in the user's situation. In particular, in situations of high urgency, the correction score exceeds the history score, enabling immediate adaptation of the presentation structure. In general information processing systems that require dynamic management of information node structures, a reconfiguration process based on the score fluctuation amount ΔS is applied, realizing a configuration that can adapt in real time to structural changes and interest fluctuations within the network. This can be developed as a reconfiguration algorithm that can operate not only on phrase networks but also on structured node information groups in general.

[0113] The link score evaluation and reconfiguration process between phrase nodes in this invention is applicable not only to dynamic optimization of phrase networks but also to other network systems requiring information transmission efficiency and selective routing control. For example, quantum communication control technology involves the configuration and reconfiguration of quantum entangled link networks. The link score evaluation, score fluctuation monitoring ΔS, reconfiguration triggering based on a threshold ΔT, and stable link fixation processes based on this invention can also be applied to the control of such quantum networks. In particular, the score evaluation and reconfiguration algorithm of this invention can be utilized in processes such as quantum bit routing and transmission optimization through reconfiguration, and evaluation of the stability of entangled structures, and future development into a communication structure optimization framework is also within the realm of possibility. This configuration is essentially similar to the present invention in that it provides dynamic network control in response to history and fluctuations. Therefore, this invention can be deployed as a structural control platform for a wide range of information transmission networks, including quantum communication. [Industrial Applicability]

[0114] As described above, the information processing device 100 and information processing method according to the present invention are configured to dynamically reconstruct the presentation structure based on the user's selection history and external information, and can be widely used industrially in the following fields: Dynamic presentation control of disaster prevention related phrase networks in disaster information systems ·History-dependent reconfiguration of product recommendation structure on e-commerce sites - Support for generating co-occurrence networks of trending words on SNS and detecting new words Optimizing response structure based on selection history in call center FAQ systems - Prompt generation support in large-scale language models (LLM) - Presentation of related structures of phrases In these application fields, the present invention can be applied as a structure control means for improving the contextual consistency of presented information and responsiveness to environmental changes. [Explanation of symbols]

[0115] 100 Information processing device 110 Input Interface 120 History Acquisition Unit 130 Score Evaluation Section 140 Structural reconstruction part 150 structural database 160 output interface 170 Feedback Line 210 processors 220 memory 230 Storage 240 communication interface 250 Input Devices 260 Output Device

Claims

1. a score evaluation unit that calculates link scores between phrase nodes; a structure reconstructing unit that reconstructs a presentation structure based on the link score; a storage unit that stores the reconstructed phrase network structure, the score evaluation unit calculates a history score, which is a link score calculated based on a co-occurrence frequency, a temporal proximity, and a selection order agreement, and is calculated based on a selection history of the phrase node; The structure reconfiguration unit If the calculated link score is below a predetermined threshold T, the corresponding link is deleted. If the link score of the link exceeds the threshold T again, the corresponding link is reconnected; The information processing device, wherein the history score is calculated by the following formula: History score R = w1 × f_co + w2 × f_time + w3 × f_order (where f_co is the score related to co-occurrence frequency, f_time is the score related to temporal proximity, and f_order is the score related to the degree of matching in selection order. w1, w2, and w3 are weighting coefficients corresponding to each evaluation index, where w1 + w2 + w3 = 1 and 0≦wi≦1 (i=1, 2, 3).)

2. 2. The information processing device according to claim 1, Weighting factor w １ , w ２ , w ３ is dynamically updated based on user attributes or external information.

3. 3. The information processing device according to claim 1, The score evaluation unit calculates a corrected score based on the reliability and urgency assigned to the external information, and calculates the link score by weighting and combining the corrected score with the history score.

4. 2. The information processing device according to claim 1, The structure reconfiguration unit deletes links whose link scores fall below a predetermined static threshold T.

5. 2. The information processing device according to claim 1, The information processing device wherein the storage unit holds history information relating to a link when the link is deleted.

6. 6. The information processing device according to claim 5, The storage unit stores a score history of a link that was previously deleted, and when a selection history related to the link is accumulated again, the storage unit reevaluates the link score and reconnects the link.

7. 3. The information processing device according to claim 1, the structural reconfiguration unit includes a configuration for designating a link as a fixed link when the same link structure continues to be reconfigured after executing the structural reconfiguration process a certain number of times, and the designation of a fixed link is conditioned on the condition that a fluctuation amount ΔS of the link score falls below a predetermined difference threshold ΔT a predetermined number of times in succession, and the ΔT is calculated based on the interquartile range (IQR) or Kullback-Leibler divergence for the time series distribution of the link score.

8. 3. The information processing device according to claim 1, the structure reconfiguration unit includes a configuration for executing reconfiguration as a condition-linked process when a variation amount ΔS of the link score exceeds a predetermined difference threshold ΔT; The ΔT is calculated using an interquartile range (IQR) or Kullback-Leibler divergence based on the distribution of score fluctuation history, If the ΔS is less than the ΔT, the information processing device is configured to exclude the link from the targets for reconfiguration.

9. An information processing method in which a link structure between phrase nodes is optimized by an information processing device, comprising: The information processing device, calculating a score corresponding to each link based on the selection history of the phrase node; acquiring external information for correcting the calculated link scores, and correcting the link scores of the phrase nodes related to the external information; based on the corrected link score, deleting links below a predetermined threshold and retaining or newly generating links equal to or above the predetermined threshold; Run An information processing method, wherein the step of retaining or newly generating includes a process of deleting or excluding the link from targets for reconnection based on a variation amount ΔS of the link score.

10. A computer program causing a computer to execute the information processing method according to claim 9.

Images