Life support system, life support device, and life support method

Abstract

To provide a method for supporting behavioral changes based on a subject's individual lifestyle at a low cost.SOLUTION: A life support apparatus 10 is provided with a text processing unit 130 for converting sensor information detected by a sensor group 40 into text based on a text rule 120 read from a storage unit 12, a word map generation unit 131 for arranging each word of text converted by the text processing unit 130 in a word map 122 which is a vector space having two or more dimensions, and a word map transition feature extracting unit 132 for outputting a proposal screen capable of comparing a first word map generated from words of a text converted from first sensor information acquired from the sensor group 40 with a second word map generated from words of a text converted from second sensor information acquired from the sensor group 40.SELECTED DRAWING: Figure 2

Description

[Technical Field]

[0001] The present invention relates to a life support system, a life support device, and a life support method. [Background technology]

[0002] A behavior change support system has been proposed in which a computer supports behavior change, such as discipline to change inappropriate behavior in children in child rearing. Patent Document 1 describes that the system determines the response method to be presented to the responder from response method information that indicates the response method for each behavior for each combination of the behavior, the trigger for the behavior, and the result of the behavior. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Japanese Patent Application Publication No. 2019-133638 Summary of the Invention [Problem to be solved by the invention]

[0004] To support behavioral change, we are considering a system that provides information to care recipients and other recipients on how to improve their lifestyles, in order to create a foundation for living that emphasizes eco-friendliness and well-being (health and happiness). In order to propose such a lifestyle, it is effective to understand the recipient's current lifestyle, including their habits, and to provide detailed and wide-ranging suggestions on which behaviors to continue and which behaviors to improve. Furthermore, because each recipient has their own unique lifestyle, personalized advice tailored to their lifestyle is more persuasive than a one-size-fits-all health manual.

[0005] However, conventional behavior change support systems such as those described in Patent Document 1 are limited to supporting the improvement of a single inappropriate behavior, such as "stealing a younger brother's toy," and are not suitable for use in proposing individual lifestyles. Furthermore, the method described in Patent Document 1 requires manual advance input of a rule database containing information on how to respond to inappropriate behaviors, which increases the cost of preparing for behavior change.

[0006] Therefore, the main object of the present invention is to provide support for behavioral change based on the individual lifestyle of each subject at low cost. [Means for solving the problem]

[0007] In order to solve the above problems, the life support system of the present invention has the following features. The present invention provides a life support system having a group of sensors that acquires sensor information about a subject whose lifestyle is to be measured, and a life support device that presents a proposal screen to the subject that proposes a lifestyle based on the acquired sensor information, The life support device comprises: a text conversion processing unit that converts the sensor information acquired by the sensor group into text based on a text conversion rule read from a storage unit; a word map generation unit that arranges each word of the text converted by the text conversion processing unit in a word map that is a vector space of two or more dimensions; The system is characterized by comprising a word map comparison unit that outputs the proposal screen that allows a comparison between a first word map generated from words in text converted from first sensor information acquired from the group of sensors and a second word map generated from words in text converted from second sensor information acquired from the group of sensors. Other means will be described later. [Effects of the Invention]

[0008] According to the present invention, it is possible to realize support for behavioral change based on the individual lifestyle of a subject at low cost. [Brief explanation of the drawings]

[0009] [Figure 1] 1 is a configuration diagram of a life support system according to an embodiment of the present invention. [Figure 2] 1 is a configuration diagram of a life support device according to an embodiment of the present invention. [Figure 3] 10 is a flowchart illustrating a text conversion process of the life support device according to the present embodiment. [Figure 4] 4 is a flowchart illustrating a word map generation process for a life support device, which is executed after FIG. 3 according to this embodiment. [Figure 5] 1 is a configuration diagram of a life support device according to a first embodiment. [Figure 6] 10 is a table illustrating an example of sensor information acquired by a sensor group according to the first embodiment. [Figure 7] 10 is a table showing an example of a text conversion rule according to the first embodiment. [Figure 8] 10 is a flowchart showing an example of a text conversion rule according to the first embodiment. [Figure 9] 10 is a flowchart showing an example of a text conversion rule according to the first embodiment. [Figure 10] 10 is a flowchart showing an example of a text conversion rule according to the first embodiment. [Figure 11] FIG. 1 is an explanatory diagram showing an example of a behavioral corpus according to the first embodiment. [Figure 12] FIG. 10 is a screen diagram showing an example of a word map according to the first embodiment. [Figure 13] FIG. 13 is a screen diagram in which action route information is added to the word map of the screen diagram of FIG. 12 relating to the first embodiment. [Figure 14] FIG. 10 is a configuration diagram of a life support device according to a second embodiment. [Figure 15] 10A and 10B are screen views showing conversion functions and inverse conversion functions between word maps according to the second embodiment. [Figure 16] FIG. 10 is a configuration diagram of a life support device according to a third embodiment. [Figure 17] FIG. 10 is a screenshot of a word map summarizing past opinions regarding Example 3. [Figure 18] FIG. 10 is a screenshot of a word map summarizing current opinions regarding Example 3. [Figure 19] FIG. 10 is a configuration diagram of a life support device according to a fourth embodiment. [Figure 20] FIG. 10 is an explanatory diagram showing an example of a behavioral corpus according to the fourth embodiment. [Figure 21] 13 is a flowchart showing details of a target word recommendation generation process according to the fourth embodiment. [Figure 22] 10 is an example of a word map related to the fourth embodiment. [Figure 23] FIG. 13 is a screen diagram showing an example of an output screen of a recommendation editing unit according to the fourth embodiment. [Figure 24] FIG. 2 is a hardware configuration diagram of each device of the life support system according to the present embodiment. DETAILED DESCRIPTION OF THE INVENTION

[0010] 1 is a configuration diagram of a life support system 100. The life support system 100 is configured such that a life support device 10, a terminal device 30 for inputting and outputting information, and a group of sensors 40 are connected via a network 20. The life support device 10 includes an input / output unit 11, a storage unit 12, and a calculation unit 13. The input / output unit 11 exchanges input / output data with the terminal device 30 and the sensor group 40. The storage unit 12 stores a database that is held in advance for calculations and data extracted by the calculation unit 13. The calculation unit 13 executes the functions of the life support device 10.

[0011] The terminal device 30 may be in the form of a PC, a smartphone, a smart watch, a smart speaker, or the like, but the type of the terminal device 30 is not limited to these. The terminal device 30 includes a presentation unit 31 and an input unit 32 . The presentation unit 31 presents to the service provider or the service user the information output by the life support device 10. The input unit 32 is used by the service provider or the service user to input information about the service user's behavior, state, and surrounding environment.

[0012] Sensor group 40 collects information about the behavior and state of the subject (a service user of life support device 10 and related parties) whose lifestyle is to be measured, as well as information about the subject's surrounding environment (such as the home). Sensor group 40 includes at least one of camera 41A, microphone 42A, vital sign sensor 43A, human presence sensor 44A, door sensor 45A, acceleration sensor 46A, gyro sensor 47A, illuminance sensor 48A, temperature and humidity sensor 41B, millimeter wave sensor 42B, radio wave sensor 43B, depth sensor 44B, atmospheric pressure sensor 45B, noise sensor 46B, vibration sensor 47B, odor sensor 48B, and home appliance sensor 49B.

[0013] Vital sensor 43A is worn by, for example, an elderly person living in a senior housing facility. Human presence sensor 44A, door sensor 45A, illuminance sensor 48A, and home appliance sensor 49B are installed in, for example, the senior housing facility. Furthermore, some or all of the sensor group 40 may be included in the terminal device 30. For example, if the terminal device 30 is a smartphone, information from the acceleration sensor 46A, the gyro sensor 47A, the GPS, and the like can be output to the life supporting device 10.

[0014] FIG. 2 is a configuration diagram of the life supporting device 10. As shown in FIG. The input / output unit 11 includes an information input unit 110 , an information output unit 111 , and a recommendation input unit 112 . The information input unit 110 receives input of sensor information acquired by the sensor group 40, sensor information mounted on the terminal device 30, and text information input via the terminal device 30 by the service provider or service user. The information output unit 111 outputs to the terminal device 30 information such as the word map 122 generated by the calculation unit 13, the extracted word map transition characteristics (details will be described later with reference to FIG. 4), and edited recommendations. The recommendation input unit 112 receives input of a recommendation sentence to be edited.

[0015] The storage unit 12 stores text conversion rules 120, a behavioral corpus 121, a word map 122, attribute information 123, word norms 124, and word embeddings 125. The text conversion rules 120 store rules for converting the behavior, situation, and surrounding environment of a service user into text information from the sensor information input from the sensor group 40. For example, it is a list of text corresponding to the results of class classification, time series analysis, arithmetic operations, comparison operations, logical operations, voice recognition, image recognition, etc. on the sensor information. The behavior corpus 121 stores, as a database, text information indicating the behavior or state of a service user (or their associates), or text information indicating the surrounding environment. Each sentence of the text includes information on the "content of the behavior (Do)" or the "content of the state (Be)," and at least one piece of information related to it: "When," "Where," "Who," "What," or "With Whom."

[0016] The word map 122 is data that is mapped based on the similarity of the semantics of the words used in the behavior corpus 121. The semantics of each word are vectorized using a count-based method (a method for reducing the dimension of a word co-occurrence matrix) or an inference-based method for acquiring word distributed representations 125. The proximity or distance between words in the word map 122 is determined by the magnitude of the inner product of the unit vector representation of each word. As a result, words with a high degree of semantic similarity are placed close to each other, and words with a low degree of semantic similarity are placed farther away.

[0017] The word map 122 may include not only the actions and states of the service user, but also names of places and rooms in the home or office, furniture and home appliances and their operating states, names and nicknames of the service user and related parties, time periods such as morning and night, days of the week, etc. Furthermore, the words in the word map 122 may be clustered according to the degree of their similarity. For example, places, actions, furniture, home appliances, people, time periods, etc. belonging to a cluster containing words such as "meal" or "dining" may be collectively classified as a "meal cluster," while places, actions, furniture, home appliances, people's names, time periods, etc. belonging to a cluster containing words such as "sleep" or "bedroom" may be collectively classified as a "sleep cluster." The number of clusters may be determined using methods such as the Elbow method or the Silhouette method.

[0018] The attribute information 123 is data labeled with the attributes of individuals or groups that are the subject of sensing, such as the level of care required for elderly people. The word norm 124 is the length of the word vector derived when vectorizing the words included in the behavior corpus 121. In the vectorized representation of a word, the more frequently a word appears in a specific context, the larger the norm is likely to be compared to other words. Therefore, words with large vector norms can be extracted as words that more clearly express the behavioral tendencies, thought patterns, habits, etc. of service users. The word embeddings 125 are generated using an inference-based method such as Word2Vec, which converts words in the behavior corpus 121 into numerical vectors to understand their meaning.

[0019] The calculation unit 13 includes a text conversion processing unit 130, a word map generation unit 131, a word map transition feature extraction unit (word map comparison unit) 132, a specific information filtering unit 133, a word norm analysis and extraction unit 134, a related word extraction unit 135, and a recommendation editing unit 136. The text processing unit 130 converts the sensor information into text by applying the algorithm shown below to the sensor information. Classification algorithm by the classification unit 1300. · Time series analysis algorithm by the time series analysis unit 1301. Arithmetic operation algorithm by the arithmetic operation unit 1302. The comparison and logic operation unit 1303 performs a comparison operation algorithm and a logic operation algorithm. A speech recognition algorithm by the speech recognition unit 1304. · Image recognition algorithm by the image recognition unit 1305.

[0020] The text conversion by the text conversion processor 130 may use the processing results of a single type of sensor information, or may use information from two or more types of sensors. For example, it is conceivable to combine a front door sensor with a directional motion sensor to classify and determine whether to generate text for "going out" or "coming home." Furthermore, text conversion may be performed based on text conversion rules 120, which are rules for converting information linked to the sensor information or its classification results.

[0021] The word map generation unit 131 vectorizes the words in the behavior corpus 121 by natural language processing operations such as the aforementioned Word2Vec, and generates a word map 122 in which the vectorized words are arranged in a vector space of two or more dimensions. The word map generation unit 131 may generate the word map using all the text in the behavior corpus 121, or may generate the word map using text generated over a certain period of time, such as the behavior corpus for the most recent month. The word map generation unit 131 may generate the word map 122 in two or three dimensions that are highly human-readable, or may generate a word map 122 having a four- or more-dimensional vector representation. The number of dimensions of the word map 122 is preferably a number of dimensions that allows for appropriate calculation of the similarity between words arranged in the word map 122 or clustering between the arranged words. Furthermore, when arranging and visualizing four- or more-dimensional vector representations of words in the word map 122, the word map generation unit 131 may use t-distributed Stochastic Neighbor Embedding (t-SNE) or PCA (Principal Component Analysis) to reduce the number of vector dimensions to two or three, thereby creating a representation that is highly human-readable.

[0022] The word map transition feature extraction unit 132 extracts conversion functions and adjustment features for mutual conversion between word maps 122 that change over time and are generated from a specific individual or group, and different word maps 122 generated from different individuals or groups. The specific information filtering unit 133 filters specific information such as personal information contained in the behavior corpus to generate an anonymous behavior corpus.

[0023] The word norm analysis and extraction unit 134 analyzes the norms of words included in the behavior corpus as the word norms 124 . The related word extraction unit 135 performs morphological analysis on the input recommendation and extracts the words that make up the recommendation. Then, it extracts related words of the recommendation for the service user from the positions on the map of the words that make up the recommendation sentence. The recommendation editing unit 136 edits the recommendation sentence using the related words extracted by the related word extraction unit 135.

[0024] FIG. 3 is a flowchart illustrating the text conversion process of the life supporting device 10. As shown in FIG. In S11, the information input unit 110 determines whether to acquire sensor information from the sensor group 40 as information regarding the behavior and status of service users and service-related persons, and information regarding the surrounding environment, or whether to accept input of text information from the terminal device 30 (S12). When text information is input from the terminal device 30 (S11, text), the text conversion processing unit 130 converts the result of the input of the text information (S12) into a behavioral corpus 121 as shown in Figure 11 and stores it in the memory unit 12 (S13).

[0025] When sensor information is acquired from the sensor group 40 (S11, sensor), in S20, the information input unit 110 branches as follows depending on the type of acquired sensor information: The text conversion processing unit 130 converts the acquired sensor information into text by each process at the branch destination. If the sensor information is time series data (S20, time series), the classification unit 1300 converts the sensor information into text by classifying the value of the timestamp associated with the sensor based on the text conversion rule 120 (S21). If the sensor information is a vital sensor value or the like (S20, vital, etc.), the sensor information is converted into text by a time series analysis process performed by the time series analysis unit 1301, an arithmetic operation process performed by the arithmetic operation unit 1302, or a logical operation process performed by the comparison / logical operation unit 1303 for the vital sensor value (S22, details in FIG. 8).

[0026] If the sensor information is a human presence sensor value or the like (S20, human presence, etc.), the comparison and logical operation unit 1303 converts the human behavior into text by performing a logical operation on the human presence sensor value and the values ​​of various sensors (door sensor 45A, vital signs sensor 43A, illuminance sensor 48A, home appliance sensor 49B) (S23, details in Figures 9 and 10). If the sensor information is voice data or the like (S20, voice), the voice recognition unit 1304 converts the content of the speaker's speech acquired from the microphone 42A into text (S24). If the sensor information is image data or the like (S20, image), the image recognition unit 1305 converts the speaker's non-verbal information (emotions, gestures, etc.) acquired from the image of the camera 41A into text (S25). The text conversion processing unit 130 also converts the text converted in S21 to S25 into the behavior corpus 121 and stores it in the storage unit 12 (S13).

[0027] FIG. 4 is a flowchart illustrating the word map generation process of the life supporting device 10, which is executed after the process of FIG. In S31, when the process of Example 2 (process of generating a group map) described later is performed using pre-input settings (S31, Yes), the specific information filtering unit 133 protects privacy by filtering specific information such as personal information from the behavior corpus 121 (S32). The group map is a word map 122 generated from group data of an unspecified number of people. Group data is data of individuals or data of family members in the same household. According to the second embodiment, a common word map 122 can be extracted from data of an unspecified large number of people who have the same attribute (for example, the same level of care required) indicated by the attribute information 123. On the other hand, in the first embodiment described later, the word map 122 is generated from individual data.

[0028] In S41, the word map generation unit 131 generates a word map 122 from the behavior corpus 121 of S13 or S31. The word map generation unit 131 periodically executes its function every time a certain number or more of new texts are added to the behavior corpus 121, such as once a week or once every few months. In S42, the word map transition feature extraction unit 132 extracts "word map transition features" that explain the time-series transitions of the word maps 122 generated in S41. Word map transition features are features that indicate how words are arranged among multiple word maps 122 and how clusters that group those words have changed. Note that the transitions among multiple word maps 122 may be transitions over time by the same person or by different people.

[0029] In S51, if the processing of Example 3 described below (processing to summarize opinions from discussions) is performed using the settings entered in advance (S51, Yes), the word norm analysis and extraction unit 134 analyzes the norms of word vectors for words included in the behavior corpus 121 and extracts keywords by sorting the words by the magnitude of the norm (S52). Then, the word map generating unit 131 generates a word map 122 (opinion map) that shows the results of collecting opinions based on the keywords extracted in S52. This opinion map reflects the common understanding and majority opinions from the free descriptions of an unspecified large group in an easy-to-read format.

[0030] In S61, if the processing of Example 4 described below (processing to supplement the content of recommendations based on the target word) is performed using the settings entered in advance (S61, Yes), the recommendation editing unit 136 creates recommendations based on the similarity between the words in the word map 122 and the target word (S62, details in Figure 21).

[0031] In S43, the word map transition feature extraction unit 132 displays the word map 122 before the transition and the word map 122 after the transition side by side. The user can understand the word map transition features by comparing the two word maps 122. Alternatively, the word map transition feature extraction unit 132 may compare the two word maps 122 and mechanically extract the word map transition features. In S44, the terminal device 30 generates a proposal for a service to be provided, such as a nursing care service, based on the word map transition features extracted in S42. The proposal for a service to be provided may be a proposal for selecting a new service to use, or a proposal for selecting how to adjust a service that is already being used. In the process of S44, the content of the proposal may be input into the terminal device 30 by the service user or service provider, or may be mechanically generated by a robot or home appliance that provides the service. [Example]

[0032] In the first embodiment, a case where the life support system 100 is applied to the lifestyle habits of an individual elderly person living in a housing for the elderly with services (hereinafter referred to as "serviced housing") will be illustrated. The life support system 100 includes a sensor group 40 that acquires sensor information about a subject whose lifestyle is to be measured, and a life support device 10 that presents the subject with a proposal screen that suggests a lifestyle based on the acquired sensor information.

[0033] Fig. 5 is a configuration diagram of the life support device 10 of Example 1. The life support device 10 of Fig. 5 is a partial extract of the life support device 10 of Fig. 2, and specifically has the following components. The input / output unit 11 includes an information input unit 110 and an information output unit 111 . The storage unit 12 includes text conversion rules 120, a behavior corpus 121, and a word map 122. The calculation unit 13 includes a text processing unit 130, a word map generation unit 131, and a word map transition feature extraction unit 132. The text conversion processing unit 130 includes a classification unit 1300, a time series analysis unit 1301, an arithmetic operation unit 1302, and a comparison / logic operation unit 1303.

[0034] The life supporting device 10 of FIG. 5 mainly includes the following components. A text conversion processing unit 130 converts the sensor information acquired by the sensor group 40 into text based on the text conversion rule 120 read from the storage unit 12. A word map generating unit 131 arranges each word of the text converted by the text conversion processing unit 130 in a word map 122, which is a vector space of two or more dimensions. A word map transition feature extraction unit 132 that outputs a proposal screen that allows a comparison between a first word map generated from words in text converted from first sensor information acquired from the sensor group 40 and a second word map generated from words in text converted from second sensor information acquired from the sensor group 40.

[0035] FIG. 6 is a table showing an example of sensor information acquired by the sensor group 40 of the first embodiment. This table stores various sensor information acquired from the sensor group 40, such as acceleration data acquired from the acceleration sensor 46A, for each timestamp (Time). Note that "Human detection @ entrance = OFF" indicates that the human detection sensor 44A that detects people at the entrance did not detect a person. Also, "Human detection @ living room = ON" indicates that the human detection sensor 44A that detects people in the living room detected a person. Although the sensor information in the table in FIG. 6 is stored every minute, the sampling rate is not limited to this.

[0036] FIG. 7 is a table showing an example of the text conversion rules 120 used in S21. This table shows the classification of timestamp values ​​into eight three-hour time slots defined by the Japan Meteorological Agency, similar to the "Time" text classification. In S21, the classification unit 1300 of the text processing unit 130 classifies the value of the timestamp associated with the sensor based on the text conversion rule 120.

[0037] 8 is a flowchart showing an example of the text conversion rule 120 used in S22. This flowchart is executed when the heart rate increases. The text conversion rule 120 is generated from an empirical rule such as "most people walk at a speed of 2.5 km / h or more." If the walking speed is 2.5 km / h or faster (S101, Yes), the text conversion processor 130 classifies the text as "exercise" (S102). On the other hand, if the answer is (S101, No), the text conversion processor 130 classifies the text as "abnormal heart rate" (S103). According to the text conversion rules 120 in FIG. 8, the text conversion processing unit 130 can derive text such as "exercise" or "abnormal heart rate" by logically operating the heart rate and acceleration. Alternatively, the text conversion processor 130 may generate text such as "heart rate increase" or "heart rate decrease" by time series analysis of the heart rate, or may derive text such as "walking" by arithmetic operations using an acceleration sensor or a gyro sensor.

[0038] 9 is a flowchart showing an example of the text conversion rule 120 used in S23. This flowchart is executed when the door open / close sensor value turns ON. The comparison and logic operation unit 1303 performs a logic operation on the door sensor value and the human sensor value in accordance with the flowchart of FIG. 9, thereby converting the value into text (S23). When the motion sensor 44A at the entrance of the house, which is positioned facing inward, is ON (S111, Yes), the text conversion processing unit 130 classifies the text as "going out" (S112). On the other hand, when the motion sensor 44A is ON (S111, No) and the most recent behavior corpus 121 contains "going out" (S113, Yes), the text conversion processing unit 130 classifies the text as "returning home" (S114). On the other hand, when the motion sensor 44A is ON (S113, No), the text conversion processing unit 130 classifies the text as "visiting" (S115).

[0039] 10 is a flowchart showing an example of the text conversion rule 120 used in S23. This flowchart is executed when the human sensor value is turned ON. The comparison and logical operation unit 1303 converts human behavior into text by performing logical operations on the human sensor value and the values ​​of various sensors (acceleration sensor 46A, illuminance sensor 48A, home appliance sensor 49B, etc.) according to the flowchart of Figure 10 (S23). The text conversion processing unit 130 may convert only the sensor information with the most recent timestamp into text, or may convert sensor information linked to multiple timestamps going back in time from the most recent sensor information into text.

[0040] The text conversion processing unit 130 branches the process depending on the location of the person whose motion sensor value is ON (S201). If the user is in the living room, the text is classified as "relaxing" (S202). If the user is in the dining room and moving near the dining table (acceleration @ dining table is above the threshold) (S211, Yes), the text is classified as "meal" (S212). If the user is in the bedroom, and the bedroom is dark (illuminance @ bedroom is less than the threshold) (S221, Yes), the text is classified as "Sleep" (S222). If the user is in the kitchen and the kitchen appliances are operating (the appliance sensor 49B detects that the refrigerator or microwave door has been opened or closed) (S231, Yes), the text is classified as "cooking" (S232).

[0041] FIG. 11 is an explanatory diagram showing an example of the behavior corpus 121 generated in S13. The text conversion processing unit 130 generates the behavioral corpus 121 of FIG. 11 showing sentences indicating human behavior such as "sleeping in the early hours of the morning" from a collection of texts showing behaviors such as "eating" and "sleeping" generated by the text conversion rules 120 of FIGS. 7 to 10, by the procedure exemplified below (S13). (Step 1) Determine the time label from the timestamp (When). (Step 2) Determine the location label from the human sensor value (Where). (Step 3) Determine the action content (Do) from other sensors (acceleration, heart rate, door, illuminance, home appliances). (Step 4) Subdivide the activity label based on the timestamp (for example, subdivide "meal" into "breakfast"). (Step 5) Create sentences showing the actions of the same person with the same timestamp from (Step 1) to (Step 4).

[0042] FIG. 12 is a screen diagram showing an example of the word map 122 (FIG. 2) extracted in S41. The word map generation unit 131 vectorizes the meaning of each word in the text converted by the text conversion processing unit 130 into a vector expression, calculates the similarity between each word based on the vector expression, and arranges words with greater similarity so that they are closer to each other in the word map 122.

[0043] The word map generating unit 131 generates the word map 122 shown in FIG. 12 or the like from the behavior corpus 121 shown in FIG. 11 or the like by the procedure exemplified below (S41). (Step 1) The word map generating unit 131 subdivides the text into parts of speech by machine morphological analysis. For example, the first line in Figure 11, "Sleep before dawn," is subdivided into the parts of speech "before dawn," "to," and "sleep." (Step 2) A machine learning algorithm (Word2Vec) that vectorizes the words of the subdivided parts of speech acquires embedded representations of the words, and plots the words on the word map 122 according to the embedded representations. Then, the word map generation unit 131 groups (clusters) the words based on the arrangement of the one or more plotted words and the distance between the words on the word map 122. The embedded representations of words are, for example, as follows: · Word "pre-dawn" = [0, 0.3, 0.4] · word "sleep" = [0.5, 0, 0.2]

[0044] (Step 3) The generated clusters are named by a person according to empirical rules. For example, it is decided that a person's activities at home should be classified into "sleeping-related," "relaxation-related," "going out-related," and "housework / eating-related." In the word map 122 of FIG. 12, a total of four clusters, namely, "sleeping C" (short for "Cluster"), "relaxation C," "going out C," and "housework / eating C," are generated based on the similarity between words, but it is not necessary to create clusters, and the number and types of clusters are not limited to these.

[0045] By using the above procedure, even if the same procedure is used, different input behavior corpora 121 will generate different output word maps 122. For example, in Figure 12, even if the same behavior corpus 121 of elderly people is used, differences occur in the individual words arranged and the clusters into which they are clustered in the following two word maps. A word map 201 (first word map) generated from sensor information (first sensor information) from one week ago. The first sensor information indicated that the health condition was "good" and the activity level was high. A word map 202 (second word map) generated from current sensor information (second sensor information). In the second sensor information, the health condition is "poor" and the activity level is low.

[0046] The word map transition feature extraction unit 132 generates a screen display 200 in which the word map 201 and the word map 202 are compared side by side, and displays the screen display 200 on the terminal device 30 (terminal of the senior housing provider) or the like (S43). Word map 201 includes a total of four clusters: Sleeping C, Relaxing C, Going Out C, and Housework / Eating C, while word map 202 includes a total of three clusters: Sleeping C, Relaxing C, and Housework / Eating C. In other words, by comparing word map 201 and word map 202 in screen display 200, the user can understand that the user has stopped going out because their health has deteriorated.

[0047] The word map transition feature extraction unit 132 extracts word map transition features between the previously generated word map 201 and the currently generated word map 202. The word map transition features may be words that have been added or removed from the previous word map 122, or clusters that have been added or removed from the previous word map 122. Alternatively, they may be words that have changed their position from a specific word or the cluster to which they belong since the previous word map 122. 12, words that are differences between word map 201 and word map 202 as word map transition features extracted by word map transition feature extraction unit 132 are surrounded by squares. For example, "Sleep C" in word map 201 when the health condition is good does not include the words "abnormal heart rate" or "morning." On the other hand, "Sleep C" in word map 202 when the health condition is poor does include the words "abnormal heart rate" and "morning."

[0048] The word map transition feature extraction unit 132 in the first embodiment extracts the following example word map transition features: The extracted word map transition features may also be included in the screen display 200 in S43. (Feature A) The disappearance of "outing-related clusters" (Feature B) The words "abnormal heart rate" and "morning" were added to the "sleep-related cluster." (Feature C) Words such as "before noon" and "afternoon" moved from the "going out" cluster to the "relaxation" cluster.

[0049] From the screen display 200 in S43, the provider of the assisted living facility can infer that the health condition and activity level of the elderly person in question have declined. As a result, the provider of the assisted living facility can propose the following exemplary services via the terminal device 30 based on the grasped word map transition characteristics (S44). For (Feature A), suggestions to encourage outdoor walks or participation in recreational activities at day care centers. Suggestions can also be made from words near "going out" C. For (Feature B) and (Feature C), alert the staff of the senior housing facility of "deteriorating health" and "changes in lifestyle" and suggest increasing the frequency of patrols.

[0050] FIG. 13 is a screen diagram in which action route information is added to the word map of the screen diagram of FIG. The word map transition feature extraction unit 132 generates behavioral paths 201R, 202R for the first word map 201 and the second word map 202 by connecting words that indicate time periods in the word map 201 that are close to each other in the word map 122 and in the order in which the time periods progressed, and displays the generated behavioral paths 201R, 202R in each word map. By comparing the behavioral routes 201R and 202R, the elderly person can understand that their current lifestyle (current behavior) is a nocturnal one, staying up late at night and not going out during the day, and that their lifestyle from one week ago (target behavior) was a diurnal one, going out during the day. Therefore, the elderly person can gain an awareness from the screen diagram of Figure 13 that they should change their lifestyle from a nocturnal one to a diurnal one.

[0051] Therefore, the word map transition feature extraction unit 132 generates a behavior path 201R that connects, for example, "before noon" of going out C → "in the evening" of housework and meals C → "early evening" of relaxation C → "late at night" of bedtime C in that order. Similarly, the word map transition feature extraction unit 132 generates a behavior path 202R that connects, in that order, "before noon" of relaxation C → "in the evening" of housework and meals C → "late at night" of bedtime C. As a result, an action path 201R that connects in the order of "Going out C → Housework / Meal C → Relaxation C → Sleeping C" is added to the word map 201. An action path 202R that connects in the order of "Relaxation C → Housework / Meal C → Sleeping C" is added to the word map 202. [Example]

[0052] In Example 2, instead of the word map 122 related to a specific individual as in Example 1, a word map 122 of a group having a specific attribute (here, level of care required) is generated, and word map transition characteristics are extracted between word maps 122 with different levels of care required. The following is an example in which the life support system 100 is applied to a group of elderly people in a senior housing facility.

[0053] 14 is a configuration diagram of a life supporting device 10 in Example 2. In the life supporting device 10 in FIG. 14, attribute information 123 and a specific information filtering unit 133 are added to the life supporting device 10 in FIG. The storage unit 12 further stores attribute information 123 for classifying multiple subjects. The word map generation unit 131 generates one word map 122 from sensor information related to multiple subjects having the same attribute information 123. That is, the attribute information 123 is referenced in the process of storing text in the behavior corpus by the text conversion processor 130 (S13 in FIG. 3) to sort the text into different behavior corpuses 121 for each attribute. The attribute information 123 is, for example, information on the level of care required for each individual, and data of individuals with the same level of care required is pooled into the same behavior corpus 121.

[0054] The specific information filtering unit 133 creates an anonymous behavioral corpus 121 by filtering and deleting information corresponding to personal information (such as name, subscriber number, and ID) from the behavioral corpus 121 (S32). In other words, when generating one word map 122 from sensor information related to multiple subjects, the specific information filtering unit 133 excludes from the word map 122 words corresponding to information that can identify the subjects. Then, the word map generating unit 131 generates a word map 122 for each different attribute (here, level of care required) for the behavior corpus 121 of S32 (S41).

[0055] FIG. 15 is a screen diagram showing a conversion function 213 and an inverse conversion function 214 between the word maps 211 and 212. The word map transition feature extraction unit 132 extracts conversion rules (conversion function 213 and inverse conversion function 214) for converting between word maps 122 (FIG. 2) of mutually different attribute information 123. The word map transition feature extraction unit 132 generates the other word map by applying a conversion rule to one of the first word map 211 and the second word map 212, which corresponds to the same attribute information 123 as the subject.

[0056] A word map 211 for a light level of care required and a word map 212 for a heavy level of care required are displayed side by side on screen 210 (S43). This makes it possible to suggest a lifestyle that will bring the heavy level of care required group closer to the light level of care required group. For this reason, word map transition feature extraction unit 132 derives a conversion function 213 from word map 211 to word map 212, and an inverse conversion function 214 from word map 212 to word map 211.

[0057] Specifically, the word map transition feature extraction unit 132 calculates a conversion function 213 and an inverse conversion function 214 for converting word maps of different attributes such as those shown in Figure 15 into each other based on the fact that the word map transition features of S42 are different in the arrangement tendencies of words (words indicating time periods, sleep, meals, etc.) that are common to word maps 122 of different levels of care required. In addition to the screen 210, the word map transition feature extraction unit 132 may display the specific contents of the conversion function 213 and the inverse conversion function 214 as an alert or recommendation to the elderly person or the service provider of the senior housing facility. For example, the presence or absence of the word "morning" in a meal-related cluster, and the difference between the clusters to which "before noon" and "afternoon" belong, etc., may be displayed. This makes it possible to recommend that a group requiring a high level of care have breakfast or go out before or after noon.

[0058] The conversion function 213 and the inverse conversion function 214 include, for example, the following (Conversion B1) to (Conversion B3). (Conversion B1) Changing the distance between words or clusters. For example, in FIG. 15, the distance between "morning" and "before noon" is changed between the word maps 122. (Transformation B2) Changing the number of clusters. (Conversion B3) Change of words that make up a cluster. For example, in Figure 15, "morning" is changed from "housework meal C" to "bedtime C." Furthermore, the conversion function 213 and the inverse conversion function 214 may perform (Conversion B1) to (Conversion B3) in a composite manner. In this case, the word map transition feature extraction unit 132 may adjust only the distances between specific words or clusters, or may use the distances between specific words to adjust the distances between all words or clusters.

[0059] The conversion function 213 and the inverse conversion function 214 thus derived are used, for example, in the following situations. Suppose an elderly person X with a severe level of care need enters a senior housing facility, and the word map generation unit 131 generates a word map 122 (word map X1) based on the lifestyle habits of the elderly person during the first week of admission. This word map X1, when used alone, visualizes the lifestyle habits of a person with a severe level of care need. By comparing the word map 122 with the target word map 122 (word map X2) showing how to improve the elderly person X's lifestyle habits, the elderly person X can become aware of ways to improve their lifestyle. However, since elderly person X has just entered a senior housing facility during the first week of admission, there is no past data, and therefore word map X2 cannot be created from the personal data of elderly person X.

[0060] Therefore, the word map generating unit 131 applies the inverse transformation function 214 to the word map X1 to create a modified word map X2. This word map X2 is based on the word map X1 of the elderly person X, and is therefore more persuasive to the elderly person X than if they were shown a general (group) word map Y1 with a low level of care required.

[0061] Furthermore, suppose that three months have passed since elderly person X entered the facility. This allows a word map X3 for three months after admission to be created, and as described in Example 1, it is possible to display the individual's word map X1 from the first week of admission and the individual's word map X3 three months after admission side by side. Here, the word map transition feature extraction unit 132 generates a new conversion function X4 from word map X1 to word map X3. The word map transition feature extraction unit 132 may display the result of comparing this individual conversion function X4 with the group conversion function 213 (or inverse conversion function 214) displayed in FIG. 15. This makes it possible to analyze whether the transition state of elderly person X is heading in the direction of worsening or improving the level of care required.

[0062] In S44, for example, the terminal device 30 of the insurance business operator periodically determines which attribute word map 122 the elderly person enrolled in the senior housing facility is closest to or is transitioning to, and recommends that the elderly person review the product they are enrolled in. Alternatively, the terminal device 30 can recommend that the enrollee change their behavior so that their lifestyle gradually approaches that of the word map 122 for a mild level of care required. As a result, if the word map 122 of the elderly person approaches that for a mild level of care required, the insurance premium can be reduced, and added value in the insurance business can be improved. [Example]

[0063] In Example 3, an example is given in which the life support system 100 is applied to discussions (collecting opinions and formulating improvement measures) for improving the working environment in a company's labor union. Within a company's labor union, opinions are regularly collected from union members through discussions and surveys for improving the working environment. However, the collection of diverse opinions, the extraction of major issues, and the formulation of improvement measures for those issues are often carried out by union executive committee members in addition to their regular duties, which can be a burden on employees. Therefore, in Example 3, a function is provided as a tool to support opinion gathering and the like.

[0064] FIG. 16 is a configuration diagram of a life supporting device according to the third embodiment. The life support device 10 in Fig. 16 has the same components as the life support device 10 in Fig. 5 except that the classifying unit 1300, the time series analyzing unit 1301, the arithmetic operation unit 1302, and the comparison / logical operation unit 1303 are deleted. In addition, the life support device 10 in Fig. 16 has the same components as the life support device 10 in Fig. 5 except that a voice recognizing unit 1304, an image recognizing unit 1305, a specific information filtering unit 133, a word norm analyzing / extracting unit 134, and a word norm 124 are added.

[0065] In S11, the information input unit 110 acquires information about the microphone and camera 41A in the discussion related to the association. The acquired information about the discussion is, for example, as follows: The speech recognition unit 1304 converts the content of the speaker's speech into text (S24). The image recognition unit 1305 converts non-verbal information (emotions such as happiness or anger) showing the state of the speaker when speaking from the camera 41A or the like into text (S25). Text information from questionnaires collected in writing or on the web is input from the terminal device 30 (S12). The text processing unit 130 converts the discussion information input in S24, S25, and S12 into a behavior corpus 121 and stores it in the storage unit 12 (S13).

[0066] In S52, the word norm analysis and extraction unit 134 analyzes the norms of words included in the behavioral corpus 121 in S13. Note that the more frequently a word appears in a similar context, the larger the norm is likely to be compared to other words, and the more likely it is to express a concern shared by many union members. Then, based on the analysis of the word norms 124, the word norm analysis and extraction unit 134 sorts the words by the size of the norm and extracts the top words as keywords for working environment issues. In this way, the text conversion processor 130 converts the text into text containing words that indicate the content of the meeting. Then, the word map generator 131 excludes from the word map 122 any words that indicate the content of the meeting whose vector norm is less than a predetermined value.

[0067] FIG. 17 is a screen view of a word map 221 that summarizes past opinions in the third embodiment. In S53, the word map generation unit 131 generates a word map 221 (past opinion map) that aggregates past opinions based on the keywords extracted from past discussions in S52. In the past opinion map of Fig. 17, words used in similar contexts or similar topics are placed close to each other (clustered), making it possible to classify problems facing the working environment and extract their contents. · Problem 1 C = Miscellaneous tasks, efficiency, workload, late-night work, and overtime are clustered under A: Work structure and workload issues. The second problem, C = working from home, IT tools, communication, and lack of exercise, is clustered under B: Problems associated with working from home. · Problem 3 C = Wages, bonuses, basic salary, taxes, and inflation are clustered under C: Wages and cost of living problems. · Problem 4 C = Parental leave, maternity leave, labor shortages, subsidies are clustered under D: Issues related to balancing work and childcare. The word map generation unit 131 determines the number of clusters using, for example, the Elbow method, and calculates the similarity between word vectors using, for example, the k-means method. This allows the word map generation unit 131 to derive the semantic similarity between words and cluster words that have a high semantic similarity (are close to each other on the word map 221).

[0068] FIG. 18 is a screen view of a word map 222 summarizing current opinions in the third embodiment. In S53, the word map generating unit 131 generates a word map 222 (current opinion map) that summarizes current opinions based on the keywords extracted from the current discussion in S52. Then, the word map transition feature extraction unit 132 displays a screen for comparing the past opinion map (in FIG. 17) with the current opinion map (in FIG. 18), and extracts word map transition features between the two opinion maps. This extracts improved working environment problems (e.g., "night work"), remaining problems (e.g., "lack of exercise"), newly emerged problems (e.g., "high prices"), etc. As a result, in S44, the terminal device 30 of the union executive department can plan improvement measures to be requested as company policies based on the keywords extracted in S52 and the word map transition characteristics extracted in S53. [Example]

[0069] In the fourth embodiment, a case where the life support system 100 is applied to a behavioral modification recommendation service for homes and offices will be exemplified. Fig. 19 is a configuration diagram of a life support device in Example 4. The life support device 10 in Fig. 19 is configured by adding a related word extraction unit 135, a recommendation editing unit 136, a recommendation input unit 112, and a word embedding unit 125 to the life support device 10 in Fig. 5.

[0070] FIG. 20 is an explanatory diagram illustrating an example of the behavior corpus 121 in the fourth embodiment. Compared to the behavior corpus 121 in Fig. 11, the behavior corpus 121 in Fig. 20 has a subject added to each behavior. The classification unit 1300 performs personal identification when adding a subject to text. This personal identification method may be face recognition using images captured by the camera 41A, or vital signs such as heart rate and gait, which have a higher degree of privacy compared to images.

[0071] In S41, the word map generation unit 131 generates the word map 122 based on the word embeddings 125 that indicate the meaning of each word that appears in the behavior corpus 121. Since the behavior corpus 121 is assigned a subject, actions and places that are highly relevant to a specific person are arranged in the word map 122.

[0072] FIG. 21 is a flowchart showing the details of the target word recommendation generation process (S62). In S621, the recommendation input unit 112 receives, via the terminal device 30, an input of an expression that serves as a clue for a recommendation (hereinafter, a clue sentence). The recommendation input unit 112 receives input of a cue sentence. The input cue sentence directly instructs the service user to change their behavior. For example, the cue sentence may be a behavior related to health or well-being, such as "Get some sleep," "Exercise," "Relax," "Converse," or "Take a bath." Alternatively, the cue sentence may be a behavior related to energy conservation, such as "Turn off unused lights" or "Raise the temperature setting of the air conditioner."

[0073] Alternatively, in S621, the recommendation input unit 112 may input a clue sentence calculated based on sensor data such as heart rate data measured by an elderly person wearing a wristwatch (smart watch) equipped with a sensor group 40 such as a vital sensor 43A.

[0074] FIG. 22 is an example of the word map 122 in the fourth embodiment. In S622, the related word extraction unit 135 extracts related words of the clue sentence input in S621 from the word map 122 generated in S41. The related word extraction unit 135 performs morphological analysis on the clue sentence, identifies words and their synonyms (hereinafter referred to as "target words") included in the recommendation content of the analysis result on the word map 122, and extracts related words. For example, from the clue sentence "Let's relax," the target word "relax" is identified. Then, the related word extraction unit 135 may extract related words for the target word from words (Ichiro, living room, television) in a cluster that includes the target word, as shown in Figure 22. Alternatively, if no cluster is formed on the word map 122, the related word extraction unit 135 may select words located in the vicinity of the target word.

[0075] In S623, the recommendation editing unit 136 generates a recommendation sentence by mechanically or manually editing a clue sentence from the related words extracted in S622. The recommendation editing unit 136 edits a recommendation from the related words extracted in S621 and the target word of the clue sentence. The recommendation editing unit 136 generates a recommendation sentence by applying the related words to recommendation sentence templates such as "content of action (Do)," "content of state (Be)," "when," "where," "who," "what," and "with whom." For this reason, a word dictionary such as "living room = a noun indicating a place, and is a candidate for "where" is prepared in advance. In addition, the recommendation editing unit 136 generates meaningful recommendation sentences by interpolating between the applied parts of speech with particles, auxiliary verbs, verbs, etc. The target word of the clue sentence is a candidate for the verb to be completed.

[0076] FIG. 23 is a screen diagram showing an example of an output screen of the recommendation editing unit 136 in the fourth embodiment. In S624, the recommendation editing unit 136 outputs the recommendation sentence edited in S623 to the presentation unit 31 of the terminal device 30. 23 shows an example of a recommendation message being displayed on the locked screen of a smartphone. In addition to the locked screen, the message may be displayed on a smartwatch, or may be output as a voice notification via a home robot or smart speaker.

[0077] In this way, the related word extraction unit 135 extracts the target word from the expression that serves as a clue for the input recommendation, searches for the target word from the words in the word map 122 generated by the word map generation unit 131, and extracts related words that are located near the target word in the word map 122. Then, the recommendation editing unit 136 generates a recommendation sentence that complements the input expression that serves as a clue for the recommendation, based on the target word and related words extracted by the related word extraction unit 135.

[0078] FIG. 24 is a hardware configuration diagram of each device in the life support system 100. Each device of the life support system 100 (life support device 10, terminal device 30, sensor group 40) is configured as a computer 900 having a CPU 901, RAM 902, ROM 903, HDD 904, communication I / F 905, input / output I / F 906, and media I / F 907. The communication I / F 905 is connected to an external communication device 915. The input / output I / F 906 is connected to an input / output device 916. The media I / F 907 reads and writes data from a recording medium 917. Furthermore, the CPU 901 executes a program (also called an application or an app for short) loaded into the RAM 902 to improve and control each processing unit. This program can be distributed via a communication line or recorded on a recording medium 917 such as a CD-ROM and distributed.

[0079] 24 may be integrated into one physical computer. Alternatively, computer 900 may be distributed across multiple computing resources such as cloud servers and edge servers, each of which is connected via a network. In addition, the functions of multiple devices (terminal device 30, life support device 10) may be centrally located as a single device (for example, within terminal device 30), or only the calculation unit 13 may be located on a cloud server.

[0080] The life support system 100 of each of the above-described embodiments analyzes text data of information indicating the service user's behavior, information indicating their status, and information indicating their surrounding environment as a word map 122. As a result, by deriving words that indicate the service user's lifestyle and utilizing them in the service content, it is possible to provide effective life support that is personalized according to the service user's lifestyle and its changes. The words that indicate lifestyle in the word map 122 include, for example, words that indicate behaviors, statuses, times, places, people, etc. that are highly relevant to specific behaviors or statuses.

[0081] On the other hand, conventional systems for supporting behavioral change cannot provide the same effects as the life support system 100 of this embodiment for the following reasons. · Because the rules for behavioral change are uniform for everyone, the content presented may not match the lifestyle of the service user. When encouraging people to change their behavior, direct instructions or orders are given rather than being made aware of the change, which can lead to ignorance or resistance.

[0082] Furthermore, the present invention is not limited to the above-described embodiments, and various other applications and modifications are possible without departing from the spirit of the present invention as defined in the claims. For example, the above-described embodiments provide a detailed and specific description of the configuration of the life support system 100 in order to clearly explain the present invention, and the system is not necessarily limited to having all of the components described. Furthermore, it is possible to replace part of the configuration of one embodiment with a component of another embodiment. It is also possible to add a component of another embodiment to the configuration of one embodiment. It is also possible to add, replace, or delete other components from part of the configuration of each embodiment.

[0083] Furthermore, the above-described configurations, functions, processing units, etc. may be partially or entirely realized in hardware, for example, by designing them as integrated circuits, etc. As the hardware, a broad processor device such as an FPGA (Field Programmable Gate Array) or an ASIC (Application Specific Integrated Circuit) may be used. Furthermore, each component of the life support system 100 according to the above-described embodiment may be implemented in any hardware as long as the respective hardware can transmit and receive information to and from each other via a network. Furthermore, the processing performed by a certain processing unit may be realized by a single piece of hardware, or may be realized by distributed processing using multiple pieces of hardware. [Explanation of symbols]

[0084] 10 Life Support Devices 11 Input / output section 12 Storage section 13 Arithmetic section 20 Network 30 Terminal Equipment 31 Presentation part 32 Input section 40 sensors 100 Life Support System 110 Information input section 111 Information output unit 112 Recommendation input section 120 Text Conversion Rules 121 Behavioral Corpus 122 Word Map 123 Attribute information 124 Word Norm 125 word embeddings 130 Text processing unit 131 Word Map Generation Unit 132 Word map transition feature extraction unit (word map comparison unit) 133 Specific Information Filtering Unit 134 Word Norm Analysis Extraction Unit 135 Related Word Extraction Unit 136 Recommend Editorial Department

Claims

1. A life support system including a group of sensors that acquires sensor information about a subject whose lifestyle is to be measured, and a life support device that presents a proposal screen to the subject that proposes a lifestyle based on the acquired sensor information, The life support device comprises: a text conversion processing unit that converts the sensor information acquired by the sensor group into text based on a text conversion rule read from a storage unit; a word map generation unit that arranges each word of the text converted by the text conversion processing unit in a word map that is a vector space of two or more dimensions; and a word map comparison unit that outputs the proposal screen that allows a comparison between a first word map generated from words in a text obtained by converting first sensor information acquired from the group of sensors and a second word map generated from words in a text obtained by converting second sensor information acquired from the group of sensors. Life support system.

2. The word map generation unit vectorizes the meaning of each word in the text converted by the text conversion processing unit as a vector expression, calculates the similarity between each word based on the vector expression, and arranges words in the word map so that the greater the similarity between the words, the closer they are to each other. The life support system of claim 1 .

3. The word map comparison unit generates a behavioral path by connecting words indicating time periods in the first word map and the second word map that are close to each other in position in the word map and in the order in which the time periods have progressed, and displays the generated behavioral path in each of the word maps. The life support system of claim 1 .

4. The storage unit further stores attribute information for classifying a plurality of subjects, The word map generating unit generates one word map from sensor information relating to a plurality of subjects having the same attribute information. The life support system of claim 1 .

5. The life support device further includes a specific information filtering unit, When generating one word map from sensor information relating to a plurality of subjects, the specific information filtering unit excludes words corresponding to information that can identify the subjects from the word map. The life support system of claim 4.

6. the word map comparison unit extracts a conversion rule for converting the word maps of the attribute information that are different from each other into each other; and generating the other word map by applying the conversion rule to one of the first word map and the second word map that corresponds to the same attribute information as that of the target person. The life support system of claim 4.

7. The text processing unit converts the information into text including words that indicate the content of the meeting, The word map generating unit is characterized in that it excludes from the word map any word indicating the content of the meeting whose vector norm is less than a predetermined value. The life support system of claim 1 .

8. The life support device further includes a related word extraction unit and a recommendation editing unit, the related word extraction unit extracts a target word from an expression that serves as a clue for the input recommendation, searches for the target word from words in the word map generated by the word map generation unit, and extracts related words that are located near the target word in the word map; The recommendation editing unit generates a recommendation sentence that complements an input expression that serves as a clue for the recommendation, based on the target word and related words extracted by the related word extraction unit. The life support system of claim 1 .

9. A life support device that acquires sensor information about a subject whose lifestyle is measured from a group of sensors, and presents a proposal screen to the subject that proposes a lifestyle based on the sensor information, a text conversion processing unit that converts the sensor information acquired by the sensor group into text based on a text conversion rule read from a storage unit; a word map generation unit that arranges each word of the text converted by the text conversion processing unit in a word map that is a vector space of two or more dimensions; and a word map comparison unit that outputs the proposal screen that allows a comparison between a first word map generated from words in a text obtained by converting first sensor information acquired from the group of sensors and a second word map generated from words in a text obtained by converting second sensor information acquired from the group of sensors. Life support equipment.

10. A life support device that acquires sensor information about a subject whose lifestyle is measured from a group of sensors and presents a proposal screen to the subject that proposes a lifestyle based on the sensor information includes a text processing unit, a word map generating unit, and a word map comparing unit, the text conversion processing unit converts the sensor information acquired by the sensor group into text based on a text conversion rule read from a storage unit; the word map generation unit arranges each word of the text converted by the text conversion processing unit in a word map, which is a vector space of two or more dimensions; The word map comparison unit outputs the proposal screen that allows a comparison between a first word map generated from words in a text obtained by converting first sensor information acquired from the group of sensors and a second word map generated from words in a text obtained by converting second sensor information acquired from the group of sensors. Life support methods.

Images