Information processing device, information processing method, and information processing program

Abstract

A morphological analysis unit (102) analyzes a plurality of pieces of description information each describing one or more natural-language words, and extracts a word described in each piece of description information from the pieces of description information by identifying the description position of the word in the pieces of description information. A label hierarchization processing unit (103) generates a hierarchical structure between words extracted from neighboring description positions.

Description

Information processing device, information processing method, and information processing program

[0001] The present disclosure relates to a technique for generating a hierarchical structure among words included in description information.

[0002] In systems that collect and store sensor data (such as power consumption) obtained from homes, factories, buildings, etc., information about the sensors is managed. The information managed includes static information about the sensors and dynamic time-series data measured by the sensors. Generally, the static information and time-series data are stored and managed in separate tables. Here, static information about the sensors includes a description of the sensor's installation location and what the sensor is measuring. Furthermore, what the sensor is measuring is often managed in a hierarchical structure.

[0003] For example, Patent Literature 1 discloses a method for managing sensors in a hierarchical structure, and Patent Literature 2 discloses a method for dividing label information assigned to a sensor into a description of the sensor's installation location and the measurement target by morphological analysis of the label information.

[0004] Japanese Patent Application Laid-Open No. 2006-331372 International Publication No. WO2018 / 225192

[0005] When collecting and accumulating sensor data (such as power consumption) measured in homes, factories, buildings, etc., the data is saved in a sensor list table and a sensor accumulated data table. The sensor list table has a data format such as [sensor ID, label information]. The sensor accumulated data table has a data format such as [sensor ID, measurement date and time, measurement value]. ID stands for Identifier. The label information assigned to the sensor ID is text-format label information assigned by humans for sensor management. This makes it difficult for programs to handle label information. In addition, the label information has a flat data structure. On the other hand, the sensor's measurement targets are rooms, electrical equipment, etc., and are arranged hierarchically. Converting flat-structured label information into hierarchical label information makes it possible to represent information closer to the real environment. Therefore, converting label information into hierarchical label information makes it easier to aggregate and analyze data.

[0006] Patent Document 1 discloses a method for hierarchically managing sensors. However, Patent Document 1 does not disclose a method for generating a hierarchical structure from label information. Patent Document 2 proposes a method for dividing label information using label information and a word dictionary. However, it does not disclose a method for generating a hierarchical structure from the divided label information.

[0007] The main purpose of the present disclosure is to generate a hierarchical structure from description information in which words in a natural language are described, such as label information.

[0008] The information processing device according to the present disclosure has a word extraction unit that analyzes a plurality of description information, each of which describes one or more words in a natural language, and extracts words described in each description information by identifying the description position of the word in each description information, and a hierarchical structure generation unit that generates a hierarchical structure between words extracted from adjacent description positions.

[0009] According to the present disclosure, a hierarchical structure can be generated from description information in which words in a natural language such as label information are described.

[0010] 1 is a diagram showing an example of the functional configuration of an information processing device according to embodiment 1. FIG. 2 is a diagram showing an example of a sensor list table according to embodiment 1. FIG. 3 is a diagram showing an example of a sensor accumulation data table according to embodiment 1. FIG. 4 is a diagram showing morphological analysis processing according to embodiment 1. FIG. 5 is a diagram showing label layering processing according to embodiment 1. FIG. 6 is a diagram showing an example of operation in morphological analysis processing and label layering processing according to embodiment 1. FIG. 7 is a diagram showing an example of a layered label information DB according to embodiment 1. FIG. 8 is a diagram showing an example of the hardware configuration of an information processing device according to embodiment 1. FIG. 9 is a diagram showing an example of the functional configuration of an information processing device according to embodiment 2. FIG. 10 is a diagram showing label layering processing according to embodiment 2. FIG. 11 is a diagram showing an example of a layered label information DB according to embodiment 2.

[0011] Hereinafter, embodiments will be described with reference to the drawings. In the following description of the embodiments and in the drawings, the same reference numerals denote the same or corresponding parts.

[0012] Embodiment 1. ***Description of Configuration*** Fig. 1 shows an example of the functional configuration of an information processing device 100 according to this embodiment. Also, Fig. 8 shows an example of the hardware configuration of the information processing device 100 according to this embodiment. The information processing device 100 according to this embodiment is a computer. The operation procedure of the information processing device 100 corresponds to an information processing method. Also, a program that realizes the operation of the information processing device 100 corresponds to an information processing program. First, an example of the hardware configuration of the information processing device 100 will be described with reference to Fig. 8.

[0013] The information processing device 100 includes, as hardware, a processor 901, a main storage device 902, an auxiliary storage device 903, and a communication device 904. The functions of the morphological analysis unit 102, the label layering processing unit 103, and the sensor data analysis and compilation processing unit 106 shown in FIG. 1 are realized, for example, by programs, and the programs are stored in the auxiliary storage device 903. Hereinafter, the morphological analysis unit 102, the label layering processing unit 103, and the sensor data analysis and compilation processing unit 106 are referred to as functional components. The programs that realize the functions of the functional components are loaded from the auxiliary storage device 903 to the main storage device 902. The processor 901 then executes the programs to perform the operations of the functional components in FIG. 1. FIG. 8 schematically illustrates a state in which the processor 901 is executing programs that realize the functions of the functional components in FIG. 1.

[0014] Next, an example of the functional configuration of the information processing device 100 will be described with reference to FIG.

[0015] The sensor list table 101 stores static information related to sensors (hereinafter referred to as sensor static information). The sensor list table 101 is, for example, a CSV file table. The sensor list table 101 may also be an XML file table. The sensor list table 101 may also be a binary file table. Furthermore, the sensor list table 101 may also be a relational database table.

[0016] The morphological analysis unit 102 divides the label information included in the sensor list table 101 into words through morphological analysis processing. The sensor list table 101 contains a plurality of pieces of label information. Each piece of label information describes one or more words in a natural language that represent the attributes of the sensor (installation location, measurement target, etc.). The morphological analysis unit 102 analyzes each piece of label information and extracts the words described in each piece of label information from each piece of label information by identifying the position of the word. The label information corresponds to description information. The morphological analysis unit 102 also corresponds to a word extraction unit. The processing performed by the morphological analysis unit 102 also corresponds to word extraction processing.

[0017] The label layering processing unit 103 generates a hierarchical structure from the label information divided into words. More specifically, the label layering processing unit 103 generates a hierarchical structure between words extracted from adjacent description positions by the morphological analysis unit 102. The label layering processing unit 103 corresponds to a hierarchical structure generation unit. The processing performed by the label layering processing unit 103 corresponds to a hierarchical structure generation process.

[0018] The hierarchical label information DB 104 stores hierarchical label information. The hierarchical label information DB 104 is, for example, a CSV file. Alternatively, the hierarchical label information DB 104 may be an XML file or a binary file. Alternatively, the hierarchical label information DB 104 may be any of a table in a relational database, a document in a document database, a key-value in a key-value database, and a node and an edge in a graph database.

[0019] The sensor accumulated data table 105 stores time-series data of sensor measurement values. The sensor accumulated data table 105 is, for example, a CSV file table. The sensor accumulated data table 105 may also be an XML file table. The sensor accumulated data table 105 may also be a binary file table. Furthermore, the sensor accumulated data table 105 may also be a relational database table.

[0020] The sensor data analysis and aggregation processing unit 106 analyzes and aggregates the sensor data using the information in the hierarchical label information DB 104 and the information in the sensor accumulated data table 105. The sensor data analysis and aggregation processing unit 106 analyzes and aggregates the sensor data, thereby providing various services and solutions.

[0021] FIG. 2 shows an example of the sensor list table 101 .

[0022] Sensor static information is stored in the sensor list table 101. In the example of FIG. 2, sensor IDs and label information are stored in the sensor list table 101 as sensor static information. Other information that may be stored includes the sensor model number, installation date and time, installation location, size, and service life. When the sensor list table 101 is stored in the table format of a relational database, the sensor ID serves as the primary key. The sensor ID is an identifier for uniquely identifying a sensor. In the example of FIG. 2, the sensor ID is a numeric value, but the sensor ID may also be in text format.

[0023] Label information is text-formatted character string information assigned by humans for the purpose of managing sensors. The label information contains natural language words that describe the sensor's installation location and the sensor's measurement target. In the example in Figure 2, character strings (words) are separated by spaces. However, character strings (words) do not need to be separated.

[0024] FIG. 3 shows an example of the sensor accumulation data table 105 .

[0025] The sensor accumulated data table 105 stores time-series data measured by the sensor. In the example of FIG. 3, the sensor ID, the measurement date and time, and the measurement value are stored. Other information such as measurement accuracy and measurement error may also be stored. When the sensor accumulated data table 105 is stored in a relational database table format, the sensor ID and the measurement date and time form a composite primary key, and the sensor ID forms a foreign key. The sensor ID is an identifier for uniquely identifying the sensor. In the example of FIG. 3, the sensor ID is a number, but the sensor ID may also be in text format. The sensor ID in the sensor accumulated data table 105 uses the same value as the sensor ID in the sensor list table 101. The measurement date and time is the date and time when the sensor performed the measurement. In addition to the date and time, time zone information may also be stored. The measurement value is the value measured by the sensor. In the example of FIG. 3, the measurement value is a number, but it may also be binary data or text data.

[0026] ***Explanation of Operation*** Next, an outline of the operation of the information processing device 100 according to this embodiment will be described.

[0027] First, the user saves sensor static information in a sensor list table 101. The sensor measurement values ​​are saved as time-series data in a sensor accumulated data table 105. The user runs a morphological analysis unit 102 and a label layering processing unit 103 to generate layered label information prior to analyzing the sensor accumulated data. Thereafter, a sensor data analysis and aggregation processing unit 106 performs analysis and aggregation processing of the sensor data.

[0028] FIG. 4 shows the morphological analysis process performed by the morphological analysis unit 102.

[0029] In step S41, the morphological analysis unit 102 determines whether or not morphological analysis has been performed on all label information included in the sensor list table 101. If morphological analysis has been performed on all label information (YES), the morphological analysis unit 102 ends the morphological analysis process. On the other hand, if there is label information for which morphological analysis has not been performed (NO), the process proceeds to step S42. By repeating steps S41, S42, and S43, the morphological analysis unit 102 performs the processes of steps S42 and S43 on all label information.

[0030] In step S42, the morphological analysis unit 102 acquires unprocessed label information from the sensor list table 101.

[0031] In step S43, the morphological analysis unit 102 performs morphological analysis on the label information acquired in step S42 and divides the label information into meaningful words. The morphological analysis unit 102 can perform morphological analysis using known technology (e.g., open-source morphological analysis software MeCab, etc.). In step S43, the morphological analysis unit 102 identifies the description position of each word in the label information and divides the label information into words. In the label information for sensor ID: 53 in FIG. 2, the morphological analysis unit 102 identifies "Instrument 1" as being written first. The morphological analysis unit 102 also identifies "(A Circuit)" as being written second. The morphological analysis unit 102 also identifies "1L-1" as being written third. The morphological analysis unit 102 also identifies "Air Conditioner_W" as being written fourth. After completing the process of step S43, the morphological analysis unit 102 marks the label information as having undergone morphological analysis processing, and returns to step S41.

[0032] FIG. 5 shows the label layering process performed by the label layering processor 103. As shown in FIG.

[0033] In step S51, the label layering processing unit 103 determines whether or not the processing has been performed on all strings of words segmented by the morphological analysis unit 102. If the processing has been performed on all strings (YES), the processing proceeds to step S53. On the other hand, if there are strings that have not been processed (NO), the processing proceeds to step S52.

[0034] The words segmented by the morphological analysis unit 102 are composed of multiple words for each piece of label information. For the sake of explanation, each piece of label information is represented as one row, and each segmented word is represented as one column. That is, for each piece of label information, the first segmented word is represented as the first column, the second segmented word is represented as the second column, and so on. The label layering processing unit 103 repeats steps S51 and S52 to perform the processing of step S52 for all columns.

[0035] In step S52, the label layering processor 103 performs a process of combining duplicate words in the target string into the same word. More specifically, the label layering processor 103 keeps only the first word that appears in the same string (words extracted from the same description position), and removes the same word that appears in the same string from the second time onwards. After completing the process in step S52, the label layering processor 103 marks the target string as processed and returns to step S51.

[0036] In step S53, it is determined whether the hierarchical structure generation process (S54, S55, S56) has been performed on all columns. If the hierarchical structure generation process has been performed on all columns (YES), the label layering processing unit 103 ends the label layering process. On the other hand, if there is a column for which the hierarchical structure generation process has not been performed (NO), the process proceeds to step S54.

[0037] In step S54, the label layering processor 103 compares the number of words in the nth column (A) with the number of words in the (n+1)th column (B). If the comparison result is "(A) < (B)," the process proceeds to step S55. On the other hand, if "(A) ≥ (B)," the process proceeds to step S56. Here, the nth column is a variable value ranging from the first column to "(the maximum value of the number of columns - 1)" when one piece of label information is represented as one row and one segmented word is represented as one column. The label layering processor 103 changes the value of n by 1 from 1 to "(the maximum value of the number of columns - 1)." By repeating steps S53 to S56, the label layering processor 103 executes the process of step S55 or step S56 for all columns.

[0038] In step S55, the label layering processor 103 generates a hierarchical structure with the word in the nth column as the parent and the word in the (n+1)th column as the child. When a word in the nth column and a word in the (n+1)th column are extracted from the same label information, the label layering processor 103 generates a hierarchical structure between the word in the nth column and the word in the (n+1)th column, both extracted from the same label information. That is, when a word is present in the nth column of the target row and also in the (n+1)th column, the label layering processor 103 generates a hierarchical structure with the word in the nth column as the parent and the word in the (n+1)th column as the child. On the other hand, when a word is not present in the nth column of the target row but is present in the (n+1)th column, the label layering processor 103 searches for a row above the target row in the nth column. The label layering processor 103 then generates a hierarchical structure with the first word found as the parent and the word in the (n+1)th column as the child. That is, the label layering processing unit 103 generates a layered structure between the word in the (n+1)th column and the word in the nth column of label information that is different from the label information from which the word in the (n+1)th column was extracted.

[0039] In step S56, unlike step S55, the label layering processing unit 103 does not generate a layer structure. A column for which a layer structure is not generated becomes a layer with no children. In addition, the (n+1)th column becomes a layer with no parent.

[0040] FIG. 6 shows an example of the operation of the morphological analysis process and the label hierarchical process according to the first embodiment.

[0041] Reference numeral 601 indicates the state before steps S41, S42, and S43 in Fig. 4 are executed. In the example of Fig. 6, two pieces of label information, "Instrument 1 (A circuit) 1L-1 Air Conditioner_W" and "Instrument 1 (A circuit) 2L-1 Air Conditioner_W," exist in the sensor list table 101.

[0042] Reference numeral 602 indicates the state after steps S41, S42, and S43 in Fig. 4 are executed and before steps S51 and S52 in Fig. 5 are executed. In the example of Fig. 6, "meter 1 (A circuit) 1L-1 air conditioner_W" is divided into four words, "meter 1," "(A circuit)," "1L-1," and "air conditioner_W," by the morphological analysis process (step S43). Also, "meter 1 (A circuit) 2L-1 air conditioner_W" is divided into four words, "meter 1," "(A circuit)," "2L-1," and "air conditioner_W," by the morphological analysis process (step S43).

[0043] Reference numeral 603 indicates the state after steps S51 and S52 in FIG. 5 are executed and before steps S53, S54, S55, and S56 in FIG. 5 are executed. In the example of FIG. 6, step S52 is first performed on the first column of reference numeral 602. The first column contains "Kei-ki 1" and "Kei-ki 1," and the same word appears multiple times. In this case, "Kei-ki 1" in the first row, which is the first word to appear, is retained, and "Kei-ki 1" in the second row, which is the word that appears the second or subsequent time, is removed. The second column of reference numeral 602 is processed in the same way as the first column. The third column of reference numeral 602 does not contain the same word multiple times, so no processing is performed. The fourth column of reference numeral 602 is processed in the same way as the first column.

[0044] Reference numeral 604 indicates the state after steps S53, S54, S55 and S56 of FIG. 5 have been executed.

[0045] In the example of FIG. 6, the first column of code 603 is first subjected to a determination in step S54. Since there is one type of word in the first column, (A) = 1. Since there is one type of word in the second column, (B) = 1. Since "(A) ≥ (B)", the process proceeds to step S56. In step S56, a hierarchical structure is not generated. Therefore, the first column becomes a layer with no children. Furthermore, since there is no parent node, the first column has a hierarchical structure with only one layer, with no parents or children.

[0046] Next, step S54 is performed to determine the second column of code 603. Since there is one type of word in the second column, (A) = 1. Since there are two types of words in the third column, (B) = 2. Since "(A) < (B)", processing proceeds to step S55. In step S55, a hierarchical structure is generated in which the words in the second column are parents and the words in the third column are children. In other words, the word in the second column, "(A circuit)", is the parent, and the two words in the third column, "1L-1" and "2L-1", are children.

[0047] Next, step S54 is performed on the third column of code 603. Since there are two types of words in the third column, (A) = 2. Since word 1 is a type in the fourth column, (B) = 1. Since "(A) ≥ (B)", the process proceeds to step S56. In step S56, a hierarchical structure is not generated. Therefore, the third column becomes a hierarchy with no children.

[0048] Since the process has been executed for all columns, the determination in step S53 becomes YES, and the process ends.

[0049] FIG. 7 shows an example of the layered label information DB 104 according to the first embodiment.

[0050] Fig. 7 shows an example in which the hierarchical processed label information 604 shown in Fig. 6 is represented as nodes and edges of a graph database and stored in the hierarchical label information DB 104. Although the example in Fig. 7 shows a graph database, the hierarchical structure may also be stored in the form of a relational database, a document database, or the like. Also, in the example in Fig. 7, there is a distinction between start points and end points of edges, but the distinction between start points and end points does not have to be made.

[0051] The word "Instrument 1" in the first column has become a single layer through the processing of step S56. Therefore, no edges are generated between words for the word "Instrument 1" in the first column. As a relationship with the sensor nodes, an edge is generated between the "Instrument 1" node and the "Sensor ID: 53" node. Also, an edge is generated between the "Instrument 1" node and the "Sensor ID: 54" node.

[0052] Between the word "(A Circuit)" in the second column and the words "1L-1" and "2L-1" in the third column, a hierarchical structure is generated by the processing of step S55, with the second column as the parent and the third column as the child. Therefore, an edge is generated between the "(A Circuit)" node and the "1L-1" node. Also, an edge is generated between the "(A Circuit)" node and the "2L-1" node. Furthermore, as a relationship with the sensor node, an edge is generated between the "1L-1" node and the "sensor ID: 53" node. Also, an edge is generated between the "2L-1" node and the "sensor ID: 54" node.

[0053] The word "air conditioner_W" in the fourth column has become a single hierarchical layer as a result of the processing in step S56. Therefore, no edges are generated between words for the word "air conditioner_W" in the fourth column. As a relationship with the sensor node, an edge is generated between the "air conditioner_W" node and the "sensor ID: 53" node. Also, an edge is generated between the "air conditioner_W" node and the "sensor ID: 54" node.

[0054] In this way, the hierarchical label information DB 104 stores the relationship between the hierarchical label information data and the sensor ID.

[0055] ***Explanation of Effects of the Embodiment*** As described above, according to this embodiment, a hierarchical structure between words can be generated from label information. This makes it possible to express information that is closer to the real environment, facilitating data aggregation and analysis. As a result, it becomes possible to provide a variety of services and solutions.

[0056] In the present embodiment, the label information representing the attributes of a sensor has been used as an example for explanation. However, the information for which the information processing device 100 according to the present embodiment can generate a hierarchical structure between words is not limited to label information. As long as the description information contains words in one or more natural languages, the information processing device 100 according to the present embodiment can generate a hierarchical structure between words included in the description information.

[0057] The procedure described in this embodiment is an example. Therefore, only a part of the procedure described in this embodiment may be implemented. Also, at least a part of the procedure described in this embodiment may be implemented in combination with a procedure not described in this embodiment. Also, the configuration and procedure described in this embodiment may be changed as necessary.

[0058] Embodiment 2. In recent years, the use of generative AI using large-scale language models (LLMs) for information aggregation / analysis or the construction of services / solutions has been generally considered. LLM stands for Large Language Model. AI stands for Artificial Intelligence. On the other hand, since LLMs learn publicly available information, they can answer questions about public information, but cannot answer questions that require internal company information or specific domain knowledge. Therefore, RAG has been proposed as a technology that improves answer accuracy by preparing knowledge that LLMs have not learned as external data and using it as an auxiliary. RAG stands for Retrieval-Augmented Generation. RAG includes GraphRAG, which handles graph-format databases as external data.

[0059] When using generative AI and GraphRAG for data analysis, knowledge that has not been learned by the LLM must be prepared in advance as graph-format data. However, converting knowledge into graph-format data is not an easy task.

[0060] In this embodiment, a method is described that enables the generation AI and GraphRAG to use the hierarchical structure as external knowledge.

[0061] In this embodiment, differences from embodiment 1 will be mainly described. Note that matters not described below are the same as those in embodiment 1.

[0062] *** Description of Configuration *** Fig. 9 shows an example of the functional configuration of the information processing device 100 according to this embodiment. In Fig. 9, compared to Fig. 1, a user 201, a generation AI+GraphRAG 202, and an LLM 203 are added. In this embodiment, the operation of the sensor data analysis and tabulation processing unit 106 described in the first embodiment is performed by the generation AI+GraphRAG 202.

[0063] A user 201 uses the information processing device 100. The user 201 inputs a question in natural language for information analysis to the generation AI+GraphRAG 202. The user 201 then receives the analysis result from the generation AI+GraphRAG 202.

[0064] The generation AI+GraphRAG 202 performs analysis and aggregation processing of the sensor data. The generation AI+GraphRAG 202 acquires a question from the user 201. Next, the generation AI+GraphRAG 202 acquires knowledge required for the analysis from the hierarchical label information DB 104. The generation AI+GraphRAG 202 also acquires data to be analyzed from the sensor accumulated data table 105. The generation AI+GraphRAG 202 then transmits the question, knowledge required for the analysis, and data to be analyzed to the LLM 203. Thereafter, the generation AI+GraphRAG 202 receives the analysis results from the LLM 203. Finally, the generation AI+GraphRAG 202 responds with the analysis results to the user 201.

[0065] 9, the sensor list table 101, morphological analysis unit 102, hierarchical label information DB 104, and sensor accumulated data table 105 are the same as those shown in FIG. 1. Therefore, detailed description of these will be omitted. In this embodiment, the label layering processing unit 103 generates hierarchical label information after the label layering process described in embodiment 1. The hierarchical label information corresponds to hierarchical structure relationship information.

[0066] The information processing device 100 according to this embodiment is also realized by, for example, the hardware configuration shown in FIG.

[0067] ***Explanation of Operation*** Next, an outline of the operation of the information processing device 100 according to this embodiment will be described.

[0068] In this embodiment, the user 201 also saves sensor static information in the sensor list table 101. The sensor measurement values ​​are saved as time-series data in the sensor accumulation data table 105. Furthermore, the user 201 executes the morphological analysis unit 102 and the label layering processing unit 103 to generate layered label information prior to analyzing the sensor accumulation data. Thereafter, the generation AI+GraphRAG 202 performs analysis processing and aggregation processing of the sensor data.

[0069] In this embodiment, the morphological analysis unit 102 also performs morphological analysis processing in accordance with the procedure shown in Fig. 4. The operation of the morphological analysis unit 102 according to this embodiment is the same as that of the first embodiment, and therefore a description thereof will be omitted.

[0070] FIG. 10 shows the label layering process performed by the label layering processor 103 according to the second embodiment.

[0071] Steps S51 to S56 are the same as those shown in Fig. 5, and therefore a description thereof will be omitted.

[0072] In step S95, the label layering processing unit 103 generates hierarchical label information. More specifically, the label layering processing unit 103 analyzes the hierarchical structure generated in step S55 and extracts one or more parent-child relationships between words in the hierarchical structure. Furthermore, the label layering processing unit 103 determines the attributes of the parent-child relationship for each parent-child relationship. The label layering processing unit 103 then generates hierarchical label information indicating one or more parent-child relationships between words in the hierarchical structure together with the attributes of each parent-child relationship. The label layering processing unit 103 also sets a parent-child relationship between any word included in the hierarchical structure and a sensor ID. The label layering processing unit 103 then determines the attributes of the parent-child relationship between a word included in the hierarchical structure and a sensor ID using at least one of a word in a parent-child relationship with the sensor ID and a word in a parent-child relationship with the sensor ID. The label layering processing unit 103 then generates hierarchical label information indicating the parent-child relationship between the word included in the hierarchical structure and the sensor ID together with the attributes of the parent-child relationship.

[0073] The label layering processing unit 103 generates layered label information, for example, in the following procedure. First, the label layering processing unit 103 generates one node for each parent word. Then, the label layering processing unit 103 sets a type attribute indicating the type of the node and a name attribute indicating the name of the node to that node. The label layering processing unit 103 may also set other attributes related to the node to that node. The label layering processing unit 103 sets the column name "Label Information" in the sensor list table 101 as the value of the type attribute. The label layering processing unit 103 also sets the word in the nth column as is as the value of the name attribute. Next, the label layering processing unit 103 similarly generates one node for each child word. Then, the label layering processing unit 103 sets a type attribute indicating the type of the node and a name attribute indicating the name of the node to that node. The label layering processing unit 103 sets the value of the type attribute to the column name "label information" in the sensor list table 101. The label layering processing unit 103 also sets the value of the name attribute to the word in the nth column as is.

[0074] The label layering processing unit 103 also generates a node for the sensor ID. Then, the label layering processing unit 103 sets a type attribute indicating the type of the node and a name attribute indicating the name of the node to the node. The label layering processing unit 103 sets the column name "sensor ID" in the sensor list table 101 as the value of the type attribute. The label layering processing unit 103 also sets the value of "sensor ID" as is as the value of the name attribute. The sensor ID node generated in this way will be referred to as a sensor ID node below.

[0075] Next, the label layering processing unit 103 generates an edge between the parent node and the child node. The direction of the edge may be either parent → child or child → parent. However, the label layering processing unit 103 unifies the direction of all edges within the layered label information. The label layering processing unit 103 sets a type attribute to the edge, which indicates the type of the edge. The label layering processing unit 103 may also set other attributes related to the edge to the edge. The label layering processing unit 103 sets a value indicating the parent-child relationship, such as "parent" or "child", to the value of the type attribute.

[0076] The label layering processing unit 103 also establishes a parent-child relationship between the words included in the hierarchical structure generated in step S55 and the sensor ID. Specifically, if a type:label information node does not have any other type:label information child nodes, the label layering processing unit 103 establishes a parent-child relationship between the type:label information node and the sensor ID node. In other words, the label layering processing unit 103 adds an edge between a type:label information node that does not have any type:label information child nodes and the sensor ID node, thereby establishing a parent-child relationship between these nodes. The label layering processing unit 103 treats the type:label information node as a parent node and the sensor ID node as a child node. The edge direction may be either from the label information node (parent) to the sensor ID node (child) or from the sensor ID node (child) to the label information node (parent). However, the label layering processing unit 103 unifies the direction of all edges within the hierarchical label information. The label layering processing unit 103 may also determine the attribute of the parent-child relationship between a label information node (parent node of the sensor ID) and a sensor ID using words in the label information node (parent node of the sensor ID) that has a parent-child relationship with the sensor ID. The label layering processing unit 103 may also determine the attribute of the parent-child relationship between a label information node (parent node of the sensor ID) and a sensor ID using words in the label information node (parent node of the sensor ID) that has a parent-child relationship with the sensor ID node. The label layering processing unit 103 may also determine the attribute of the parent-child relationship between a label information node (parent node of the sensor ID) and a sensor ID using words in the parent node of the sensor ID or the parent node of the parent node of the sensor ID and another node listed in the same column number (n-th column) in the sensor list table 101. The label layering processing unit 103 may also determine the attribute of the parent-child relationship using words in any of the above-mentioned nodes using an external dictionary or an LLM. That is, the label layering processing unit 103 may use the words of the parent nodes of the sensor ID to determine the attributes of the parent-child relationship using an external dictionary or LLM.The label layering processing unit 103 may also determine the attribute of the parent-child relationship using an external dictionary or an LLM by using the words of the parent node of the parent node of the sensor ID. The label layering processing unit 103 may also determine the attribute of the parent-child relationship using an external dictionary or an LLM by using the words of another node described in the same column number (nth column) in the sensor list table 101. Then, the label layering processing unit 103 adds a value (type attribute) corresponding to the determined attribute to the edge.

[0077] FIG. 11 shows an example of the layered label information DB 104 according to the second embodiment.

[0078] Fig. 11 shows an example in which the hierarchical processed label information 604 shown in Fig. 6 is represented as nodes and edges of a graph database and stored in the hierarchical label information DB 104. Although the example in Fig. 11 shows a graph database, the hierarchical structure may also be stored in the form of a relational database, a document database, or the like. Also, in the example in Fig. 11, there is a distinction between start points and end points of edges, but the start points and end points do not need to be distinguished.

[0079] For the word "Instrument 1" in the first column, the label layering processing unit 103 generates a node with two attributes, "type: label information, name: Instrument 1." Furthermore, for the word "(Circuit A)" in the second column, the label layering processing unit 103 generates a node with two attributes, "type: label information, name: (Circuit A)." Furthermore, for the word "1L-1" in the third column, the label layering processing unit 103 generates a node with two attributes, "type: label information, name: 1L-1." Similarly, for the word "2L-1" in the third column, the label layering processing unit 103 generates a node with two attributes, "type: label information, name: 2L-1." Furthermore, for the word "Air Conditioner_W" in the fourth column, the label layering processing unit 103 generates a node with two attributes, "type: label information, name: Air Conditioner_W." Furthermore, the label layering processing unit 103 generates a node having two attributes, "type: sensor ID, name: 53," for sensor ID: 53. Similarly, the label layering processing unit 103 generates a node having two attributes, "type: sensor ID, name: 54," for sensor ID: 54.

[0080] The word "Instrument 1" in the first column has become a single layer through the processing of step S56. Therefore, the label layering processing unit 103 does not generate edges between words for the word "Instrument 1" in the first column. In other words, the node for the word "Instrument 1" ("type: label information, name: instrument 1") does not have a child node of type: label information. Therefore, the label layering processing unit 103 sets a parent-child relationship between the node for the word "Instrument 1" and each of the two sensor ID nodes ("type: sensor ID, name: 53" and "type: sensor ID, name: 54"). Specifically, the label layering processing unit 103 generates edges between the node for the word "Instrument 1" and each of the two sensor ID nodes. Here, the label layering processing unit 103 uses the value "Instrument 1" of the name attribute of the node of the word "Instrument 1," which is the parent node of the two sensor ID nodes, and determines "Instrument" as the attribute of the parent-child relationship using an external dictionary or LLM.The label layering processing unit 103 then sets the attribute value "Instrument," which corresponds to the determined attribute of the parent-child relationship, as the type attribute value of each edge.

[0081] Furthermore, the label layering processing unit 103 sets a parent-child relationship between the node of the word "(A circuit)" in the second column ("type: label information, name: (A circuit)") and the node of the word "1L-1" in the third column ("type: label information, name: 1L-1"). Specifically, the label layering processing unit 103 generates an edge between the two nodes. Similarly, the label layering processing unit 103 sets a parent-child relationship between the node of the word "(A circuit)" in the second column and the node of the word "2L-1" in the third column ("type: label information, name: 2L-1"). Specifically, the label layering processing unit 103 generates an edge between the two nodes. Then, the label layering processing unit 103 sets "type: child" as the attribute value of the parent-child relationship for each edge.

[0082] Furthermore, the node for the word "1L-1" does not have a child node with type: label information. Therefore, the label layering processing unit 103 establishes a parent-child relationship between the node for the word "1L-1" and the sensor ID node ("type: sensor ID, name: 53"). Specifically, the label layering processing unit 103 generates an edge between the node for the word "1L-1" and the sensor ID node. Here, the label layering processing unit 103 uses the value "(A Circuit)" of the name attribute of the word "(A Circuit)," which is the parent node of the node for the word "1L-1," which is the parent node of the sensor ID node, and determines "circuit" as the attribute of the parent-child relationship using an external dictionary or LLM. Then, the label layering processing unit 103 sets the attribute value "circuit," which corresponds to the attribute of the determined parent-child relationship, as the type attribute value of the edge.

[0083] Similarly, the node for the word "2L-1" does not have a child node with type: label information. Therefore, the label layering processing unit 103 sets a parent-child relationship between the node for the word "2L-1" and the sensor ID node ("type: sensor ID, name: 54"). Specifically, the label layering processing unit 103 generates an edge between the node for the word "2L-1" and the sensor ID node. Here, the label layering processing unit 103 uses the value "(A Circuit)" of the name attribute of the word "(A Circuit)," which is the parent node of the node for the word "2L-1," which is the parent node of the sensor ID node, and determines "circuit" as the attribute of the parent-child relationship using an external dictionary or LLM. Then, the label layering processing unit 103 sets the attribute value "circuit," which corresponds to the attribute of the determined parent-child relationship, as the type attribute value of the edge.

[0084] The word "air conditioner_W" in the fourth column is in a single layer as a result of the processing in step S56. Therefore, the label layering processing unit 103 does not generate edges between words for the word "air conditioner_W" in the fourth column. In other words, the node for the word "air conditioner_W" ("type: label information, name: air conditioner_W") does not have a child node of type: label information. Therefore, the label layering processing unit 103 establishes a parent-child relationship between the node for the word "air conditioner_W" and each of the two sensor ID nodes ("type: sensor ID, name: 53" and "type: sensor ID, name: 54"). Specifically, the label layering processing unit 103 generates edges between the node for the word "air conditioner_W" and each of the two sensor ID nodes. Here, the label layering processing unit 103 uses the value "air conditioner_W" of the name attribute of the node of the word "air conditioner_W", which is the parent node of the two sensor ID nodes, and determines "equipment" as the attribute of the parent-child relationship using an external dictionary or LLM.The label layering processing unit 103 then sets the attribute value "equipment", which corresponds to the determined attribute of the parent-child relationship, as the type attribute value of each edge.

[0085] In this way, the label layering processing unit 103 sets the relationship between the layered label information data and the sensor information. The layered label information DB 104 then stores the relationship between the layered label information data and the sensor information.

[0086] ***Explanation of Effects of the Embodiment*** According to this embodiment, the generation AI and GraphRAG can use the hierarchical structure as external knowledge. Furthermore, by utilizing the generation AI and GraphRAG that use the hierarchical structure as external knowledge, more accurate data analysis and aggregation processing becomes possible. As a result, it becomes possible to provide a variety of services and solutions.

[0087] Although the first and second embodiments have been described above, these two embodiments may be implemented in combination. Alternatively, one of these two embodiments may be implemented in part. Alternatively, these two embodiments may be implemented in part in combination. Furthermore, the configurations and procedures described in these two embodiments may be modified as necessary.

[0088] *** Supplementary Explanation of Hardware Configuration *** Here, a supplementary explanation of the hardware configuration of the information processing device 100 will be provided. The processor 901 shown in FIG. 8 is an IC (Integrated Circuit) that performs processing. The processor 901 is a CPU (Central Processing Unit), a DSP (Digital Signal Processor), or the like. The main storage device 902 shown in FIG. 8 is a RAM (Random Access Memory). The auxiliary storage device 903 shown in FIG. 8 is a ROM (Read Only Memory), a flash memory, an HDD (Hard Disk Drive), or the like. The communication device 904 shown in FIG. 8 is an electronic circuit that executes data communication processing. The communication device 904 is, for example, a communication chip or a NIC (Network Interface Card).

[0089] The auxiliary storage device 903 also stores an operating system (OS). At least a portion of the OS is executed by the processor 901. While executing at least a portion of the OS, the processor 901 executes programs that implement the functions of the functional components shown in FIGS. 1 and 9. The processor 901 executes the OS to perform tasks, memory management, file management, communication control, and the like. At least one of information, data, signal values, and variable values ​​indicating the results of processing of the functional components shown in FIGS. 1 and 9 is stored in at least one of the main storage device 902, the auxiliary storage device 903, and registers and cache memory within the processor 901. The programs that implement the functions of the functional components shown in FIGS. 1 and 9 may be stored on a portable recording medium such as a magnetic disk, a flexible disk, an optical disk, a compact disk, a Blu-ray (registered trademark) disk, or a DVD. Portable recording media storing the programs that implement the functions of the functional components shown in FIGS. 1 and 9 may be distributed.

[0090] Furthermore, the "unit" of at least one of the functional components shown in FIGS. 1 and 9 may be interpreted as a "circuit," a "step," a "procedure," a "process," or a "circuitry." Furthermore, the information processing device 100 may be realized by a processing circuit. The processing circuit may be, for example, a logic IC (Integrated Circuit), a GA (Gate Array), an ASIC (Application Specific Integrated Circuit), or an FPGA (Field-Programmable Gate Array). In this case, the functional components shown in FIGS. 1 and 9 are each realized as part of the processing circuit. In this specification, the generic concept of a processor and a processing circuit is referred to as a "processing circuitry." In other words, a processor and a processing circuit are each specific examples of "processing circuitry."

[0091] Finally, various aspects of the present disclosure are summarized as appendices. (Appendix 1) An information processing device having: a word extraction unit that analyzes a plurality of description information, each of which describes one or more words in a natural language, and extracts words described in each description information from each description information by identifying the description position of the word in each description information; and a hierarchical structure generation unit that generates a hierarchical structure between words extracted from adjacent description positions. (Appendix 2) The information processing device according to Appendix 1, wherein the hierarchical structure generation unit eliminates duplication among multiple words extracted from the same description position, and generates a hierarchical structure between the words extracted from adjacent description positions after the duplication has been eliminated. (Appendix 3) The information processing device according to Appendix 1 or 2, wherein the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the nth description position and the word extracted from the (n+1)th description position when the number of types of words extracted from the (n+1)th description position is greater than the number of words extracted from the nth description position. (Supplementary Note 4) The information processing device according to Supplementary Note 3, wherein the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the nth description position and the word extracted from the (n+1)th description position, with the word extracted from the nth description position as a parent and the word extracted from the (n+1)th description position as a child. (Supplementary Note 5) The information processing device according to Supplementary Note 4, wherein, when the word extracted from the nth description position and the word extracted from the (n+1)th description position exist in the same description information, the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the nth description position and the word extracted from the (n+1)th description position. (Appendix 6) The information processing device described in Appendix 4, wherein the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the (n+1)th description position and the word extracted from the nth description position of description information different from the description information from which the word extracted from the (n+1)th description position was extracted, when a word extracted from the (n+1)th description position exists for any description information but a word extracted from the nth description position does not exist.(Supplementary Note 7) The information processing device according to any one of Supplements 1 to 6, wherein the word extraction unit analyzes a plurality of pieces of description information, each of which describes one or more words in a natural language that express attributes of a sensor. (Supplementary Note 8) An information processing method, wherein a computer analyzes a plurality of pieces of description information, each of which describes one or more words in a natural language, and extracts, from each piece of description information, words described in each piece of description information by identifying the description position of the word in each piece of description information, and the computer generates a hierarchical structure between the words extracted from adjacent description positions. (Supplementary Note 9) An information processing program, which causes a computer to execute a word extraction process that analyzes a plurality of pieces of description information, each of which describes one or more words in a natural language, and extracts, from each piece of description information, words described in each piece of description information by identifying the description position of the word in each piece of description information, and a hierarchical structure generation process that generates a hierarchical structure between the words extracted from adjacent description positions. (Supplementary Note 10) The information processing device described in Supplementary Note 1, wherein the hierarchical structure generation unit analyzes the hierarchical structure, extracts one or more parent-child relationships between words in the hierarchical structure, determines attributes of the parent-child relationship for each parent-child relationship, and generates hierarchical structure relationship information indicating one or more parent-child relationships between words in the hierarchical structure together with the attributes of each parent-child relationship. (Supplementary Note 11) The information processing device according to Supplementary Note 10, wherein the word extraction unit analyzes a plurality of pieces of description information, each of which describes a sensor ID that is an ID (Identifier) ​​of a sensor, and the hierarchical structure generation unit sets a parent-child relationship between any word included in the hierarchical structure and the sensor ID, and determines an attribute of the parent-child relationship between any word included in the hierarchical structure and the sensor ID using at least one of a word that has a parent-child relationship with the sensor ID and a word that has a parent-child relationship with the sensor ID, and a word that has a parent-child relationship with the word that has a parent-child relationship with the sensor ID, and generates hierarchical structure relationship information that indicates the parent-child relationship between any word included in the hierarchical structure and the sensor ID together with the attribute of the parent-child relationship between any word included in the hierarchical structure and the sensor ID. (Supplementary Note 12) The information processing device according to Supplementary Note 10, wherein the hierarchical structure generation unit generates hierarchical structure relationship information in a graph format.

[0092] 100 Information processing device, 101 Sensor list table, 102 Morphological analysis unit, 103 Label hierarchical processing unit, 104 Hierarchical label information DB, 105 Sensor accumulated data table, 106 Sensor data analysis and aggregation processing unit, 201 User, 202 Generation AI + GraphRAG, 203 LLM, 901 Processor, 902 Main memory device, 903 Auxiliary memory device, 904 Communication device.

Claims

1. An information processing device having: a word extraction unit that analyzes multiple pieces of description information, each of which describes one or more words in a natural language, and extracts words described in each piece of description information by identifying the description position of the word in each piece of description information; and a hierarchical structure generation unit that generates a hierarchical structure between words extracted from adjacent description positions.

2. The information processing device according to claim 1, wherein the hierarchical structure generation unit eliminates duplication among multiple words extracted from the same description position, and after the duplication has been eliminated, generates a hierarchical structure between words extracted from adjacent description positions.

3. An information processing device as described in claim 1 or 2, wherein the hierarchical structure generation unit generates a hierarchical structure between the words extracted from the nth description position and the words extracted from the (n+1)th description position when the number of types of words extracted from the (n+1)th description position is greater than the number of words extracted from the nth description position.

4. The information processing device according to claim 3, wherein the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the nth description position and the word extracted from the (n+1)th description position, with the word extracted from the nth description position as the parent and the word extracted from the (n+1)th description position as the child.

5. The information processing device of claim 4, wherein the hierarchical structure generation unit generates a hierarchical structure between a word extracted from the nth description position and a word extracted from the (n+1)th description position, both extracted from the same description information, when both words exist.

6. The information processing device of claim 4, wherein the hierarchical structure generation unit generates a hierarchical structure between the word extracted from the (n+1)th description position and the word extracted from the nth description position of description information different from the description information from which the word extracted from the (n+1)th description position was extracted when a word extracted from the (n+1)th description position exists for any description information but a word extracted from the nth description position does not exist.

7. An information processing device according to any one of claims 1 to 6, wherein the word extraction unit analyzes a plurality of pieces of descriptive information, each of which describes one or more words in a natural language that each represent an attribute of a sensor.

8. An information processing method in which a computer analyzes multiple pieces of description information, each of which describes one or more words in a natural language, extracts words described in each piece of description information by identifying the position of the word in each piece of description information, and generates a hierarchical structure between words extracted from adjacent positions of description.

9. An information processing program that causes a computer to execute a word extraction process that analyzes multiple pieces of description information, each of which describes one or more words in a natural language, and extracts words described in each piece of description information by identifying the position of the word in each piece of description information, and a hierarchical structure generation process that generates a hierarchical structure between words extracted from adjacent positions of description.

10. The information processing device described in claim 1, wherein the hierarchical structure generation unit analyzes the hierarchical structure, extracts one or more parent-child relationships between words in the hierarchical structure, determines the attributes of the parent-child relationship for each parent-child relationship, and generates hierarchical structure relationship information indicating one or more parent-child relationships between words in the hierarchical structure together with the attributes of each parent-child relationship.

11. The information processing device described in claim 10, wherein the word extraction unit analyzes multiple pieces of descriptive information, each of which describes a sensor ID (Identifier) ​​of a sensor; and the hierarchical structure generation unit sets a parent-child relationship between any word included in the hierarchical structure and the sensor ID, and determines attributes of the parent-child relationship between any word included in the hierarchical structure and the sensor ID using at least one of a word that has a parent-child relationship with the sensor ID and a word that has a parent-child relationship with the sensor ID, and a word that has a parent-child relationship with the word that has a parent-child relationship with the sensor ID, and generates hierarchical structure relationship information that indicates the parent-child relationship between any word included in the hierarchical structure and the sensor ID together with attributes of the parent-child relationship between any word included in the hierarchical structure and the sensor ID.

12. The information processing device according to claim 10, wherein the hierarchical structure generating unit generates hierarchical structure relationship information in a graph format.

Images