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

The information processing device addresses the challenge of setting appropriate parameter values in graph-based search by generating a model that estimates index parameters based on query-specific learning data, enhancing search accuracy and efficiency.

JP2024165298APending Publication Date: 2024-11-28LY CORP
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
JP2023081383
Authority / Receiving Office
JP · JP
Patent Type
Applications
Current Assignee / Owner
Filing Date
2023-05-17
Publication Date
2024-11-28

AI Technical Summary

Technical Problem

Conventional information search techniques using graph indexes face challenges in setting appropriate parameter values, as these values often vary with the query, making it difficult to achieve optimal search accuracy.

Method used

An information processing device that generates a model to estimate index parameter values based on a query by repeatedly varying parameter values during extraction processes and comparing results to correct search outcomes, using learning data to adjust parameter values for accurate search performance.

Benefits of technology

The solution allows for the generation of a model that effectively estimates index parameter values according to specific queries, improving search accuracy and reducing unnecessary search time by optimizing parameter settings for each query.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure 2024165298000001_ABST
    Figure 2024165298000001_ABST
Patent Text Reader

Abstract

To enable generation of a model for estimating an index parameter value according to a query.SOLUTION: An information processing device disclosed herein comprises: an acquisition unit for acquiring an index for multiple objects, a query object, and first accuracy required to perform a search; a first generation unit configured to generate a correct search result of a search for the multiple objects using the query object; and a second generation unit configured to generate learning data to be used for learning of a model for estimating a parameter value by iteratively performing extraction processing for extracting nearby objects corresponding to the query object from among the multiple objects using an index while changing the value of the parameter for the search, and associating the query object with a value of the parameter when a difference between the first accuracy and second accuracy based on comparison between a result of the extraction processing and the correct search result satisfies a given condition.SELECTED DRAWING: Figure 4
Need to check novelty before this filing date? Find Prior Art

Description

[Technical field]

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

[0002] Conventionally, there have been provided techniques for searching (retrieving) information using various indexes. For example, there has been provided a technique for generating an index, which is a graph in which nodes corresponding to a search target are connected by edges, and performing a search using the generated graph. In addition, such techniques are used, for example, for image searches. [Prior art documents] [Patent documents]

[0003] [Patent Document 1] Patent No. 6293335 [Patent Document 2] Patent No. 6300982 [Patent Document 3] Patent No. 6311000 [Non-patent literature]

[0004] [Non-Patent Document 1] Masajiro Iwasaki, "Neighborhood Search Using Approximate k-Nearest Neighbor Graph with Tree-Structured Index", Transactions of Information Processing Society of Japan, 2011 / 2, Vol. 52, No. 2. pp.817-828. Summary of the Invention [Problem to be solved by the invention]

[0005] However, there is room for improvement in the above conventional techniques. For example, in the above conventional techniques, it is necessary to set parameters that define a search range, etc., when performing a search process using an index such as a graph, but it is difficult to set appropriate values ​​for the parameters. For example, it is expected that the parameter values ​​will vary depending on the query, and it is desirable to be able to set appropriate parameter values ​​depending on the query.

[0006] The present application has been made in consideration of the above, and aims to provide an information processing device, an information processing method, and an information processing program that enable generation of a model that estimates index parameter values ​​in response to a query. [Means for solving the problem]

[0007] The information processing device according to the present application is characterized in that it includes an acquisition unit that acquires an index targeting multiple objects to be the target of a data search, a query object for a search query in the data search, and standard information indicating a first accuracy required when performing a search using the multiple objects as search targets; a first generation unit that generates a correct answer search result indicating an object that is a correct answer in a search targeting the multiple objects using the query object; and a second generation unit that generates learning data to be used for learning a model that estimates the value of the parameter by repeatedly executing an extraction process that uses the index to extract nearby objects that correspond to the query object from the multiple objects by varying values ​​of parameters related to the search, and by associating the query object with the value of the parameter when a difference between a second accuracy based on a comparison between a result of the extraction process and the correct answer search result and the first accuracy satisfies a predetermined condition. Effect of the Invention

[0008] An advantage of one aspect of the embodiment is that it is possible to generate a model that estimates values ​​of parameters of an index in response to a query. [Brief description of the drawings]

[0009] [Figure 1] FIG. 1 is a diagram illustrating an example of information processing according to an embodiment. [Diagram 2] FIG. 2 is a diagram illustrating an example of information processing according to the embodiment. [Diagram 3] FIG. 3 is a diagram illustrating an example of the configuration of an information processing system according to the embodiment. [Figure 4] FIG. 4 is a diagram illustrating an example of the configuration of the information processing device according to the embodiment. [Diagram 5] FIG. 5 is a diagram illustrating an example of a model information storage unit according to the embodiment. [Figure 6] FIG. 6 is a flowchart illustrating an example of information processing according to the embodiment. [Figure 7] FIG. 7 is a flowchart illustrating an example of information processing according to the embodiment. [Figure 8] FIG. 8 is a diagram illustrating an example of a hardware configuration. DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS

[0010] Hereinafter, the information processing device, the information processing method, and the information processing program according to the present application will be described in detail with reference to the drawings. Note that the information processing device, the information processing method, and the information processing program according to the present application are not limited to the embodiments. In addition, the same parts in the following embodiments are denoted by the same reference numerals, and duplicated descriptions are omitted.

[0011] (Embodiment) [1. Information Processing] An example of information processing according to the embodiment will be described below with reference to Figs. 1 and 2. Figs. 1 and 2 are diagrams showing an example of information processing according to the embodiment. Specifically, Fig. 1 shows an example of processing (generation processing) for generating data (also referred to as "learning data") for learning a model according to the embodiment. Also, Fig. 2 is a diagram showing an example of search processing based on inference processing according to the embodiment. Below, a description will be given in the order of a generation processing of learning data for learning an estimation model that estimates parameter values ​​of an index used in an approximate neighborhood search (hereinafter also referred to as "neighborhood search"), a learning processing of the estimation model using the generated learning data, and a search processing of neighborhood search based on inference processing using the learned estimation model.

[0012] In the following example, a graph index (also simply called a "graph") in which multiple objects to be searched for are connected by edges will be described as an example of an index used in a neighborhood search, but the index is not limited to a graph. In other words, the graph shown below is merely an example of an index, and the index is not limited to a graph, and may be any index such as a tree-structured index, a binary index, or a direct product quantization index, as long as it is an index used in a neighborhood search.

[0013] In the following example, parameter "ε" (sometimes simply referred to as "ε"), which is a search range coefficient (see FIG. 2) in a graph search, is described as an example of a parameter whose value is to be estimated, but parameters whose values ​​are to be estimated are not limited to ε. In other words, ε shown below is merely an example of a parameter, and the parameter is not limited to ε and may be any parameter as long as it is a parameter set in a proximity search using an index.

[0014] [1-1. Example of generation process] First, an example of a process for generating learning data will be described with reference to Fig. 1. Note that Fig. 1 describes an example in which an information processing device 100 (see Fig. 4) selects an object (also called a "query object") to be used as a search query during the generation process from a group of objects (also called a "dataset") used for learning. Note that the query object may be selected from a data set other than the data set. In this case, in Fig. 1, the information processing device 100 acquires a query object from, for example, a query object group DS2, which is a data set other than the data set DS1.

[0015] In the example of FIG. 1, the information processing device 100 generates learning data for a model M1 (see FIG. 2), which is an estimation model that estimates the value of a parameter "ε", using a data set DS1 including a plurality of objects such as objects OB1, OB2, OB3, and OB101. For example, each object in the data set DS1 is vectorized, and various processes such as a search process described later are performed on the vectorized object (vector information). Note that the object may be any information as long as it can be expressed as a vector. Note that, although vector information targeting image information will be described below, the target of the vector information may be other targets such as video information and audio information.

[0016] In addition, the information used by the information processing device 100 is not limited to vectors, and may be any type of information as long as it is information capable of expressing the similarity of each object. For example, the information processing device 100 may use graph information in which the object is graph-structured using predetermined data or values ​​corresponding to each object. For example, the information processing device 100 may use graph information in which the object is graph-structured using predetermined numerical values ​​(for example, binary values ​​or hexadecimal values) generated from each object. For example, instead of vectors, data of any type may be used as long as the distance (similarity) between data is defined. For example, the information processing device 100 acquires the data from another external device such as the information providing device 50 (see FIG. 3). In addition, when the data set DS1 has already been acquired, the information processing device 100 may acquire the data set DS1 from, for example, the storage unit 120 (see FIG. 4).

[0017] First, the information processing device 100 acquires a query object (step S1). In FIG. 1, the information processing device 100 selects a query object from a given data set DS1. The information processing device 100 randomly selects a query object from the data set DS1. This allows the information processing device 100 to set an object not included in the data set as a query object. For example, the information processing device 100 selects an object OB500 or the like from the data set DS1 as a query object. Thus, in the example of FIG. 1, the information processing device 100 selects the object OB500 as a query object QE1 as shown in the query object list RQ1. The information processing device 100 generates a data group excluding the object selected as the query object from the data set DS1 as a search target data group TD1.

[0018] Then, the information processing device 100 generates a correct search result (step S2). The information processing device 100 generates a correct search result when searching the search target data group TD1 using each object selected as a query object as shown in the query object list RQ1. In FIG. 1, the information processing device 100 generates a correct search result when searching the search target data group TD1 using the object OB500 selected as the query object QE1 as a query. For example, the information processing device 100 generates a correct search result indicating a result of extracting k (k is an arbitrary number) objects (nodes) as nearby objects (nearby nodes) for the query object QE1 by linearly searching the search target data group TD1.

[0019] In Fig. 1, the information processing device 100 generates a correct answer search result RR1 corresponding to a query object QE1. As shown in the correct answer search result RR1, the approximate correct answer information corresponding to the query object QE1 has No "1", that is, the nearest node is the node corresponding to the object OB101. Also, the approximate correct answer information corresponding to the query object QE1 has No "k", that is, the farthest node (farthest object) is the node corresponding to the object OB55. The information processing device 100 uses the correct answer search result RR1 to perform processes such as accuracy calculation, which will be described later.

[0020] The generation of the correct search result is not limited to linear search, and may be performed by any method. For example, the information processing device 100 may generate the correct search result by using the graph GR11 generated in step S3. In this case, the information processing device 100 may perform step S3 before step S2. For example, the information processing device 100 may generate the correct search result corresponding to the query object QE1 by increasing the value of the parameter "ε" by a predetermined value or more and performing the processing shown in FIG. 7 using the graph GR11. Note that the above is an example, and the information processing device 100 may generate the correct search result by any processing.

[0021] Here, the concept of the parameter "ε" will be briefly explained. In the graph GR11 of FIG. 2, a "circle" with a "Q" written inside conceptually indicates a query, and conceptually illustrates ranges AR1 and AR2 corresponding to the query. The range AR1 conceptually indicates a range within a radius r centered on the query, and the range AR2 indicates a range within a radius r (1+ε) obtained by expanding the radius r centered on the query by ε. In this way, ε is used as a parameter for varying (expanding) the search range. For example, the information processing device 100 performs a process of extracting nearby objects by applying the parameter "ε", and the details of the process using the parameter "ε" will be explained in FIG. 7.

[0022] Then, the information processing device 100 generates a graph (step S3). The information processing device 100 generates a graph for the dataset DS1. In FIG. 1, the information processing device 100 generates a graph for a group of objects (search target data group TD1) excluding objects (object OB500, etc.) selected as query objects in the dataset DS1. In FIG. 1, the information processing device 100 generates a graph GR11 for the search target data group TD1. The information processing device 100 generates the graph GR11 by appropriately using various techniques related to graph generation. Here, the graph GR11 will be described.

[0023] Graph GR11 shown in Fig. 1 indicates graph information in which nodes corresponding to each object included in data group TD1 to be searched are connected by directed edges. Note that graph information such as that shown in graph GR11 in Fig. 1 is not limited to being generated by the information processing device 100, and the information processing device 100 may obtain graph information such as that shown in graph GR11 in Fig. 1 from another external device such as the information providing device 50.

[0024] In addition, the directed edge here means an edge that can trace data only in one direction. In the following, the source of tracing by the edge, i.e., the node that is the starting point, is referred to as the reference source, and the destination of tracing by the edge, i.e., the node that is the end point, is referred to as the reference destination. For example, a directed edge that connects a specific node "A" to a specific node "B" indicates an edge that has the reference source as node "A" and the reference destination as node "B". Note that the edges connecting each node are not limited to directed edges, and may be various edges. For example, the edges connecting each node may be edges that have no direction connecting the nodes. For example, the edges connecting each node may be edges that can reference each other. For example, all the edges connecting each node may be undirected edges (bidirectional edges).

[0025] For example, an edge with node "A" as the reference source is called an output edge of node "A". Also, for example, an edge with node "B" as the reference destination is called an input edge of node "B". That is, the output edge and the input edge here are the difference between which of the two nodes connected by a directed edge is regarded as the center, and one directed edge becomes an output edge and an input edge. That is, the output edge and the input edge are relative concepts, and when a directed edge is regarded as the center, it becomes an output edge, and when a directed edge is regarded as the center, it becomes an input edge. In this embodiment, the edges are directed edges such as output edges and input edges, so that hereinafter, the directed edge may be simply referred to as an "edge". Also, each node here corresponds to each object. For example, each of a plurality of local features extracted from an image may be an object. Also, for example, various data in which the distance between objects is defined may be an object.

[0026] 1 includes many objects (nodes) in the data set DS1, but only a portion of them is illustrated in the drawing. For example, the information processing device 100 generates graph information including a plurality of nodes (vectors) such as nodes N1 to N3, N43, and N53, as shown in the graph GR11 in Fig. 1. In the example of Fig. 1, for ease of explanation, five nodes are illustrated to explain the outline of the process, but the graph GR11 includes nodes in the number corresponding to the number of objects in the data set DS1.

[0027] In the example of FIG. 1, for example, the information processing device 100 generates the graph GR11 such that a predetermined number or more of output edges are connected to other edges from each node corresponding to each object. The information processing device 100 generates the graph GR1 such that each node in the graph GR11 is connected to a predetermined number of edges (output edges) from the node closest to the node. For example, the predetermined number may be various values ​​such as 2, 5, 10, 100, etc., depending on the purpose, use, etc. For example, when the predetermined number is 2, output edges are connected from the node N1 to the node closest to the node N1 and two nodes second closest to the node N1. Note that the distance as an index indicating the similarity may be any distance as long as it is applicable as a distance between vectors (N-dimensional vectors), and various distances such as Euclidean distance and Mahalanobis distance may be used. For example, the distance may be any information as long as it reflects the similarity between two objects, and may be information on an angle such as cosine similarity.

[0028] Also, when "node N* (* is an arbitrary number)" is written like this, it indicates that the node is identified by the node ID "N*". For example, when "node N1" is written, the node is identified by the node ID "N1".

[0029] In addition, in the graph GR11 in FIG. 1, the node N1 is connected to the edge E2, which is a directed edge toward the node N53. That is, the node N1 is connected to the node N53 by the edge E2. When "edge E* (* is an arbitrary numerical value)" is written in this way, it indicates that the edge is identified by the edge ID "E*". For example, when "edge E31" is written, the edge is identified by the edge ID "E31". For example, it is possible to trace from the node N1 to the node N53 by the edge E2, which is connected with the node N1 as the reference source and the node N53 as the reference destination. In this case, the edge E2, which is a directed edge, becomes an output edge when it is identified with the node N1 as the center, and becomes an input edge when it is identified with the node N53 as the center.

[0030] In other words, when viewed from the node N1 side, the edge E2, which is a directed edge, is an edge with an arrow pointing from itself to another edge, i.e., an outward edge, and when viewed from the node N53 side, it is an edge with an arrow pointing toward itself, i.e., an inward edge. That is, the output edge here can be read as an outward edge, and the input edge can be read as an inward edge. Although not shown in FIG. 1, the node N53 may be connected to a directed edge (edge ​​E531) toward the node N1. In this way, the edge E531, which is an output edge from the node N53, may be connected to the node N1. In this case, two edges are connected between the node N1 and the node N53: the edge E2, which is a directed edge from the node N1 toward the node N53, and the edge E531, which is a directed edge from the node N53 toward the node N1.

[0031] Graph GR11 in Fig. 1 may be a Euclidean space. Graph GR11 shown in Fig. 1 is a conceptual diagram for explaining distances between vectors, and is a multidimensional space. For example, graph GR11 shown in Fig. 1 is illustrated in a two-dimensional form in order to illustrate it on a plane, but it is a multidimensional space, such as 100 dimensions or 1000 dimensions. Vector data corresponding to each node may be an N-dimensional real-valued vector.

[0032] Also, in the example of FIG. 1, in graph GR11, "node N* (* is an arbitrary numerical value)" is omitted as appropriate, and each node is represented by adding the value of "*" of "node N*" within the "circle" corresponding to it. That is, the "*" in the "node N*" portion corresponds to the matching node. For example, the "circle" in the upper left corner of graph GR11 with "43" added inside it corresponds to the node (node ​​N43) identified by node ID "N43".

[0033] Here, the distance between each node indicates the similarity of the nodes (image information), and the closer the distance, the more similar the nodes are. In this embodiment, the distance between each node in the graph GR11 is taken as the similarity between the corresponding objects. For example, the similarity of the image information corresponding to each node is assumed to be mapped as the distance between the nodes in the graph GR11. For example, the similarity between the concepts corresponding to each node is assumed to be mapped to the distance between the nodes. Here, in the example of FIG. 1, the similarity between objects whose distance between each node in the graph GR11 is short is high, and the similarity between objects whose distance between each node in the graph GR11 is long is low.

[0034] For example, in the graph GR11 in FIG. 1, the node N43 and the node N2 are close to each other, that is, the distance is short (close). Therefore, the object corresponding to the node N43 and the object corresponding to the node N2 have a high similarity. Also, in the graph GR11 in FIG. 1, the node N43 and the node N53 are far away, that is, the distance is long (far). Therefore, the object corresponding to the node N43 and the object corresponding to the node N53 have a low similarity. Note that the above is an example, and the information processing device 100 may generate a graph using various conditions. For example, the information processing device 100 may generate the graph GR11 such that a predetermined number or more of input edges are connected from each node corresponding to each object to other edges.

[0035] Then, the information processing device 100 generates learning data (step S4). The information processing device 100 acquires reference information SA1 indicating the accuracy (also referred to as "first accuracy") required when performing a search with the dataset DS1 as the search target. For example, the information processing device 100 acquires reference information SA1 indicating the first accuracy A1 from the storage unit 120. The accuracy in FIG. 1 will be described as an example of a recall required when performing a search with the dataset DS1 as the search target. Note that the accuracy is not limited to the recall, and may be various information as long as the accuracy of the search process can be measured. The first accuracy A1 in FIG. 1 indicates the value of the recall required when performing a search with the dataset DS1 as the search target.

[0036] As shown in the target parameter information TP1, the information processing device 100 generates learning data by adjusting ε, which is a parameter, for the purpose. For example, the information processing device 100 repeatedly executes an extraction process for extracting nearby objects corresponding to a query object from a search target data group TD1 using a graph GR11 while varying the value of ε, and generates learning data to be used for learning a model M1 that estimates the value of ε by associating the value of ε with the query object when the difference between the first accuracy A1 and the second accuracy based on a comparison between the result of the extraction process and the correct search result falls within a predetermined value.

[0037] For example, the information processing device 100 generates learning data to be used for learning the model M1 by repeatedly executing an extraction process for extracting nearby objects corresponding to a query object while varying the value of ε, by processing as shown in Fig. 7. For example, the information processing device 100 generates learning data to be used for learning the model M1 by repeatedly executing an extraction process for extracting nearby objects corresponding to a query object while gradually increasing the value of ε.

[0038] Hereinafter, a case where the value of ε is determined for the query object QE1 (object OB500) will be described as an example. For example, the information processing device 100 executes the extraction process for the query object QE1 by gradually increasing the value of the parameter "ε" from 0. For example, the information processing device 100 sets the value of the parameter "ε" for the query object QE1 to 0 and performs the process shown in FIG. 7 using the graph GR11 to generate an extraction result RS1 that is the result of the extraction process corresponding to the query object QE1. The extraction result RS1 indicates k neighboring objects when the search process is performed for the query object QE1 with the value of the parameter "ε" set to 0.

[0039] Then, the information processing device 100 calculates the accuracy (also called "second accuracy") with the set parameters based on a comparison between the result of the extraction process and the correct search result. The information processing device 100 calculates the second accuracy when the parameter "ε" is set to 0 for the query object QE1, using the extraction result RS1 and the correct search result RR1. Then, the information processing device 100 determines whether the difference between the first accuracy A1 and the second accuracy is within a predetermined value.

[0040] The information processing device 100 increases the value of the parameter "ε" when the difference between the first accuracy A1 and the second accuracy is not within a predetermined value. For example, when the difference between the first accuracy A1 and the second accuracy is not within a predetermined value, the information processing device 100 increases the value of the parameter "ε" by a predetermined value (e.g., 0.05, 0.1, etc.). Then, the information processing device 100 repeats the above-mentioned process for the query object QE1 using the increased value of the parameter "ε".

[0041] On the other hand, when the difference between the first accuracy A1 and the second accuracy falls within a predetermined value, the information processing device 100 associates the value VL1 (e.g., 0.1, 0.25, etc.) of the parameter "ε" with the query object QE1. In Fig. 1, the information processing device 100 associates the value VL1 of the parameter "ε" with the query object QE1 when the difference between the first accuracy A1 and the second accuracy falls within a predetermined value, as shown in the learning data LD1.

[0042] Then, the information processing device 100 registers information associating the value of the parameter "ε" with the query object QE1 in the learning data storage unit 121 (see FIG. 3). In FIG. 1, the information processing device 100 stores, in the learning data storage unit 121, information associating each of the query objects such as objects OB500, OB1234, etc., as shown in the learning data LD1, with the parameter value.

[0043] As described above, the information processing device 100 collects, for each query, parameter values ​​that are estimated to be appropriate when a search process is executed using that query, in association with that query. This enables the information processing device 100 to generate an estimation model that outputs parameter values ​​that are estimated to be appropriate for a query when executing a query search process. Thus, the information processing device 100 can generate a model that estimates parameter values ​​of an index according to a query.

[0044] Generally, for example, in searches such as approximate vector neighborhood searches, the search accuracy varies depending on the query. Therefore, to always obtain high search accuracy, for example, ε (search range coefficient) can be set large, but even for queries where ε is small and sufficient accuracy can be obtained, ε will be increased, which will increase the search time. In this way, if ε (search range coefficient) is set large across the board, there may be cases where the search time increases unnecessarily.

[0045] Therefore, it is desirable to estimate an appropriate ε for each query. Therefore, the information processing device 100 can generate an estimation model capable of accurate estimation by collecting learning data required for generating an estimation model that estimates the value of ε for each query.

[0046] Specifically, the information processing device 100 generates learning data for machine learning of ε by executing the following processes (1-1) to (1-5).

[0047] (1-1): Randomly split the data to be registered into registration data and query data. If data for a query exists, that data is used as the query data. (1-2): Perform a linear search using the query data to generate correct search results (1-3): Generate an index using registered data (1-4): Gradually increase ε from 0 and search using the index, then calculate the accuracy using the correct search results. (1-5): The ε data whose accuracy reaches A is set as the output data.

[0048] The information processing device 100 can generate learning data (data set) including query data (input data) that is multidimensional vector data and the ε value (output data) through the above processes (1-1) to (1-5). The information processing device 100 can generate a large number of data sets by repeating the above processes (1-1) to (1-5).

[0049] [1-2. Learning process example] From here, an example of a learning process in which the information processing device 100 learns the model M1 using the learning data generated by the above-mentioned generation process will be described. In the following, as an example of the model M1, a case in which a model that outputs an ε value corresponding to a search query when the search query is input will be described as an example. Note that the information processing device 100 may learn a classification model, which will be described later.

[0050] The information processing device 100 acquires data (learning data) used in the learning process of the model M1. For example, the information processing device 100 acquires the learning data of the model M1 from the learning data storage unit 121 (see FIG. 4). In FIG. 1, the information processing device 100 acquires, as the learning data of the model M1, information in which each of query objects such as objects OB500 and OB1234 is associated with a parameter value as shown in the learning data LD1.

[0051] The information processing device 100 uses the learning data LD1 to learn the model M1. For example, the information processing device 100 performs a learning process by a method such as backpropagation (backpropagation method) so that a value output by the model M1 approaches correct answer information (parameter value) associated with a query object input to the model M1. For example, the information processing device 100 performs a learning process so that a value output by the model M1 to which information (vector, etc.) of a query object is input approaches correct answer information associated with the store. For example, when the correct answer information is "1", the information processing device 100 performs a learning process so that when a query object to which the correct answer information is assigned is input, a value output by the model M1 approaches "1". Also, for example, when the correct answer information is "0", the information processing device 100 performs a learning process so that when a query object to which the correct answer information is assigned is input, a value output by the model M1 approaches "0".

[0052] 1, the information processing device 100 performs a learning process when an object OB500 is input, so that a value output by the model M1 approaches a value VL1 associated with the object OB500. Also, when an object OB1234 is input, the information processing device 100 performs a learning process when an object OB1234 is input, so that a value output by the model M1 approaches a value VL2 associated with the object OB1234.

[0053] For example, the information processing device 100 adjusts the value of the weight (i.e., the connection coefficient) that is taken into consideration when a value is transmitted between nodes by the learning process. In this way, the information processing device 100 learns the model M1 by a process such as backpropagation that corrects the parameters (connection coefficients) so as to reduce the error between the output in the model M1 and the correct answer information corresponding to the input. For example, the information processing device 100 generates the model M1 by performing a process such as backpropagation to minimize a predetermined loss function. This allows the information processing device 100 to perform a learning process that learns the parameters of the model M1.

[0054] The learning method of the model is not limited to the above-mentioned method, and any known technology can be applied. The generation of each model may be performed by appropriately using various conventional technologies related to machine learning. For example, the generation of the model may be performed by using a technology related to machine learning of supervised learning, such as SVM (Support Vector Machine). Also, for example, the generation of the model may be performed by using a technology related to machine learning of unsupervised learning. For example, the generation of the model may be performed by using a deep learning technology. For example, the generation of the model may be performed by appropriately using various deep learning technologies such as DNN (Deep Neural Network), RNN (Recurrent Neural Network), and CNN (Convolutional Neural Network). Note that the description regarding the generation of the model is an example, and the generation of the model may be performed by a learning method appropriately selected according to the obtainable information, etc. That is, the information processing device 100 may generate the model M1 by any method as long as the model M1 can be learned so as to output a value corresponding to the correct answer information when the information of the query object included in the learning data is input.

[0055] The above-mentioned model is merely an example, and the information processing device 100 may learn a classification model. In this case, the value of ε may be a discrete value such as 0.10, 0.12, ..., and the classification model may be generated by performing machine learning on input data and output data as a classification problem.

[0056] For example, for each query object in the learning data LD1, the information processing device 100 assigns the value of a parameter associated with the query object to one of the above-mentioned discrete values ​​(labels). For example, the information processing device 100 assigns the value of a parameter associated with the query object to the discrete value (label) that has the smallest difference from the parameter value among the above-mentioned discrete values ​​(labels), and uses the assigned discrete value (label) as the correct answer information of the query object.

[0057] For example, when a query object is input to a classification model, the information processing device 100 performs a learning process so that the classification model classifies the query object into a discrete value (label) associated with the query object. In Fig. 1, when an object OB500 is input, the information processing device 100 performs a learning process so that the classification model outputs a classification result in which the object OB500 is classified into a discrete value (label) associated with the object OB500. Also, when an object OB1234 is input, the information processing device 100 performs a learning process so that the classification model outputs a classification result in which the object OB1234 is classified into a discrete value (label) associated with the object OB1234.

[0058] As described above, the information processing device 100 can generate an estimation model that outputs parameter values ​​estimated to be appropriate for a query by using the generated learning data. Therefore, the information processing device 100 can generate a model that estimates parameter values ​​of an index in response to a query.

[0059] [1-3. Search processing example] During search processing, the information processing device 100 estimates the value of ε from the query using the learned model, and executes the process of approximate neighborhood search using the index using the estimated value of ε. This point will be described with reference to FIG.

[0060] 2, the information processing device 100 acquires a query QR11 to be used in a search process. The information processing device 100 acquires the query QR11 from a user as a search request. For example, the information processing device 100 receives the query QR11 from a terminal device 10 used by the user.

[0061] The information processing device 100 inputs the query QR11 as input information to an estimation model that estimates a parameter value (step S11-1). In FIG. 2, the information processing device 100 inputs the vector of the query QR11 as input information to a model M1. The estimation model to which the input information has been input outputs an estimated value of the parameter (step S11-2). In FIG. 2, the model M1 to which the vector of the query QR11 has been input outputs an estimated value VL11 of the parameter "ε", as shown in output information OT1.

[0062] Then, the information processing device 100 sets the value of the parameter "ε" to the estimated value VL11, and executes a search process using the graph GR11 for the query QR11 (step S12). For example, the information processing device 100 sets the value of the parameter "ε" to the estimated value VL11, and executes the search process shown in Fig. 7 to execute a search process for extracting k objects as nearby objects for the query QR11. Note that the search process executed by the information processing device 100 is not limited to the process shown in Fig. 7, and any process using an index can be adopted.

[0063] Then, the information processing device 100 provides the user with the result of the search process for the query QR11. For example, the information processing device 100 transmits information indicating the k nearby objects, which is the result of the search process for the query QR11, to the terminal device 10 used by the user.

[0064] As described above, the information processing device 100 can estimate appropriate parameter values ​​for a query by using an estimation model that outputs parameter values ​​estimated to be appropriate for the query. Therefore, the information processing device 100 can execute a search process using an index with parameters set to appropriate values.

[0065] 2. Configuration of the Information Processing System Next, a configuration example of the information processing system 1 will be described with reference to Fig. 3. Fig. 3 is a diagram showing a configuration example of the information processing system according to the embodiment. As shown in Fig. 3, the information processing system 1 includes an information processing device 100, an information providing device 50, and a terminal device 10. The information processing device 100, the information providing device 50, and the terminal device 10 are connected to each other via a predetermined network N so as to be able to communicate with each other by wire or wirelessly. Note that the information processing system 1 shown in Fig. 3 may include a plurality of information processing devices 100, a plurality of information providing devices 50, and a plurality of terminal devices 10.

[0066] The information processing device 100 is a computer that executes various information processing. The information processing device 100 is a generation device that generates data (learning data) for learning various models such as the model M1. The information processing device 100 is also a learning device that learns various models such as the model M1 using the generated learning data. The information processing device 100 is also an inference device that executes estimation processing using various models such as the learned model M1. The information processing device 100 is also a search device that executes approximate neighborhood search using an index such as a graph index. The information processing device 100 also executes search processing using a result of estimation processing using various models such as the learned model M1.

[0067] The information processing device 100 is a computer that provides a service to a user who uses a terminal device 10. The information processing device 100 provides a search service. The information processing device 100 provides a search processing result to the user. The information processing device 100 transmits the search processing result to the terminal device used by the user.

[0068] The information providing device 50 is a server device that provides various information to the information processing device 100. For example, the information providing device 50 provides an index such as a graph index to the information processing device 100. The information providing device 50 provides graph data such as the graph GR11 to the information processing device 100.

[0069] The information providing device 50 provides the information processing device 100 with information (data) that can be used for model learning. For example, the information providing device 50 transmits data that is permitted to be used for model learning to the information processing device 100 as learning data. For example, the information providing device 50 publishes public data on the Internet, and transmits the requested data to the external device, such as the information processing device 100, in response to a request from the external device. For example, the information providing device 50 receives request information requesting data from the information processing device 100, and transmits data corresponding to the request information to the information processing device 100 in response to the reception of the request information.

[0070] The terminal device 10 is an information processing device used by a user. The terminal device 10 is realized by, for example, a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. The terminal device 10 may be referred to as a user. In other words, the user may be read as the terminal device 10.

[0071] The terminal device 10 accepts operations by a user. In addition, the terminal device 10 transmits various information to the information processing device 100 and the information providing device 50 in response to the user's operations. For example, the terminal device 10 transmits request information requesting various information to the information processing device 100 and the information providing device 50. In addition, the terminal device 10 receives various information from the information processing device 100 and the information providing device 50.

[0072] The terminal device 10 transmits a search query input by a user to the information processing device 100. The terminal device 10 transmits the search query input by the user to the information processing device 100 as a search request. The terminal device 10 receives various information from the information processing device 100. For example, the terminal device 10 receives search results from the information processing device 100. For example, the terminal device 10 receives from the information processing device 100 a search process of the search query transmitted to the information processing device 100.

[0073] 3. Configuration of Information Processing Device Next, the configuration of the information processing device 100 according to the embodiment will be described with reference to Fig. 4. Fig. 4 is a diagram showing an example of the configuration of the information processing device 100 according to the embodiment. As shown in Fig. 4, the information processing device 100 has a communication unit 110, a storage unit 120, and a control unit 130. Note that the information processing device 100 may have an input unit (e.g., a keyboard, a mouse, etc.) that accepts various operations from an administrator of the information processing device 100, and a display unit (e.g., a liquid crystal display, etc.) that displays various information.

[0074] (Communication unit 110) The communication unit 110 is realized by, for example, a network interface card (NIC) etc. The communication unit 110 is connected to a network by wire or wirelessly, and transmits and receives information to and from the terminal device 10 and the information providing device 50.

[0075] (Storage unit 120) The storage unit 120 is realized by, for example, a semiconductor memory element such as a random access memory (RAM) or a flash memory, or a storage device such as a hard disk or an optical disk. As shown in FIG. 4, the storage unit 120 according to the embodiment has a learning data storage unit 121 and a model information storage unit 122.

[0076] The storage unit 120 is not limited to the learning data storage unit 121 and the model information storage unit 122, and may store various information depending on the purpose. The storage unit 120 stores various information used in the learning data generation process. For example, the storage unit 120 stores reference information indicating a first accuracy used in the learning data generation process.

[0077] The storage unit 120 stores various information related to objects. The storage unit 120 stores information on objects to be searched. The storage unit 120 stores information on indices for which parameters are to be estimated. For example, the storage unit 120 stores index information such as a graph index. For example, when the information processing device 100 provides a search service, the storage unit 120 stores various information used for the search service.

[0078] (Learning data storage unit 121) The learning data storage unit 121 according to the embodiment stores various information (learning data) related to data used for learning. The learning data storage unit 121 stores teacher data used for generating a model. The learning data storage unit 121 stores learning data information including information such as data used for input and correct answer information (correct answer value) corresponding to the data. The learning data storage unit 121 stores learning data generated by the generation unit 132.

[0079] The learning data storage unit 121 stores learning data in which a query is associated with a parameter value (correct answer value) corresponding to the query. For example, the learning data storage unit 121 stores learning data in which a query is associated with a parameter "ε" value (correct answer value) corresponding to the query.

[0080] The learning data storage unit 121 may store various kinds of information according to the purpose, not limited to the above. For example, the learning data storage unit 121 stores information indicating the type of input data in association with each input data.

[0081] (Model information storage unit 122) The model information storage unit 122 according to the embodiment stores information about a model. For example, the model information storage unit 122 stores information (model data) about a trained model (model) trained (generated) by a learning process. FIG. 5 is a diagram illustrating an example of a model information storage unit according to the embodiment. In the example shown in FIG. 5, the model information storage unit 122 includes items such as "model ID," "purpose," and "model data."

[0082] "Model ID" indicates identification information for identifying a model. "Use" indicates the use of the corresponding model. "Model Data" indicates the data of the model. Figure 5 shows an example in which conceptual information such as "MDT1" is stored in "Model Data", but in reality, various information constituting the model is included, such as information on the model configuration (network configuration) and information on parameters. For example, "Model Data" includes information including the nodes in each layer of the network, the functions employed by each node, the connection relationships between the nodes, and the connection coefficients set for the connections between the nodes.

[0083] In FIG. 5, the model (model M1) identified by the model ID "M1" has a use of "parameter value estimation", and indicates that, for example, model M1 is a model used to estimate a value of a parameter corresponding to a query. For example, model M1 indicates that model M1 is a model used to estimate an appropriate value of parameter "ε" that is set during a search process using a query. Also, it indicates that the model data of model M1 is model data MDT1.

[0084] The model information storage unit 122 may store various information according to the purpose, not limited to the above. The model such as the model M1 may be expected to be used as a program module that is a part of artificial intelligence software. The model such as the model M1 may be a program.

[0085] (Control unit 130) Returning to the explanation of Fig. 4, the control unit 130 is a controller, and is realized, for example, by a CPU (Central Processing Unit), MPU (Micro Processing Unit), etc., executing various programs (corresponding to examples of information processing programs such as learning programs and generation programs) stored in a storage device inside the information processing device 100 using a RAM as a working area. The control unit 130 is also a controller, and is realized, for example, by an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

[0086] 4, the control unit 130 has an acquisition unit 131, a generation unit 132, a learning unit 133, a processing unit 134, and a provision unit 135, and realizes or executes the functions and actions of information processing described below. Note that the internal configuration of the control unit 130 is not limited to the configuration shown in FIG. 4, and may be other configurations as long as they perform the information processing described below.

[0087] (Acquisition part 131) The acquisition unit 131 acquires various information. The acquisition unit 131 acquires various information from an external information processing device such as the terminal device 10 or the information providing device 50. The acquisition unit 131 may receive various information indicated in the learning data storage unit 121, the model information storage unit 122, etc. from the external information processing device.

[0088] The acquisition unit 131 acquires various pieces of information from the storage unit 120. The acquisition unit 131 acquires graph data from the storage unit 120. The acquisition unit 131 acquires various pieces of information from the learning data storage unit 121, the model information storage unit 122, etc. The acquisition unit 131 receives various pieces of information from an external device.

[0089] The acquisition unit 131 receives various information from the information providing device 50. For example, the acquisition unit 131 receives information used for learning from the information providing device 50. For example, the acquisition unit 131 receives correct answer search results from the information providing device 50.

[0090] The acquisition unit 131 receives various information from the terminal device 10 used by the user. The acquisition unit 131 acquires a query to be searched from the user as a search request. The acquisition unit 131 receives the query from the terminal device 10 used by the user.

[0091] The acquisition unit 131 acquires an index for multiple objects to be the target of a data search, a query object for a search query in a data search, and reference information indicating a first accuracy required when performing a search with multiple objects as search targets. The acquisition unit 131 acquires an index to be used for an approximate neighborhood search. The acquisition unit 131 acquires a graph index in which multiple objects are connected by edges as an index.

[0092] The acquisition unit 131 acquires a query object which is an object selected from a group of objects used for learning, and an index targeting a plurality of objects excluding the object selected as the query object from the group of objects. The acquisition unit 131 acquires a query object which is an object randomly selected from the group of objects.

[0093] The acquisition unit 131 acquires an index targeting a plurality of objects that are an object group used for learning, and a query object that is an object other than the object group. The acquisition unit 131 acquires a query object from a query object group different from the object group. The acquisition unit 131 acquires a target query that is a search query to be processed.

[0094] (Generation unit 132) The generation unit 132 executes a generation process to estimate various pieces of information. For example, the generation unit 132 executes the generation process based on various pieces of information acquired by the acquisition unit 131. The generation unit 132 executes the generation process based on various pieces of information stored in the storage unit 120. The generation unit 132 executes the generation process based on various pieces of information received from an external information processing device. The generation unit 132 executes a generation process to generate data (learning data) used for model learning by the learning unit 133.

[0095] The generation unit 132 repeatedly executes an extraction process that uses an index to extract nearby objects from among multiple objects that correspond to a query object by varying the values ​​of search parameters, and generates learning data to be used for training a model that estimates parameter values ​​by associating the query object with the parameter values ​​when the difference between the first accuracy and the second accuracy based on a comparison between the result of the extraction process and the correct search result satisfies a predetermined condition.

[0096] The generation unit 132 repeatedly executes the extraction process by gradually changing the parameter value in the direction of change where the accuracy of the result of the extraction process is high, and generates learning data by associating the parameter value when the difference between the second accuracy and the first accuracy satisfies a predetermined condition with the query object. The generation unit 132 repeatedly executes the extraction process by gradually increasing the parameter value, and generates learning data by associating the parameter value when the difference between the second accuracy and the first accuracy satisfies a predetermined condition with the query object.

[0097] (Learning Section 133) The learning unit 133 learns various information. The learning unit 133 generates various models by a learning process. For example, the learning unit 133 performs a learning process based on information stored in the learning data storage unit 121, etc. The learning unit 133 generates various models based on information stored in the learning data storage unit 121, etc. For example, the learning unit 133 learns a model M1 using the learning data stored in the learning data storage unit 121.

[0098] For example, the learning unit 133 executes a learning process based on various information acquired by the acquisition unit 131. For example, the learning unit 133 learns a model based on information generated by the generation unit 132. The learning unit 133 executes a learning process based on information from an external information processing device and information stored in the storage unit 120. The learning unit 133 executes a learning process based on information stored in the model information storage unit 122. The learning unit 133 stores the model generated by learning in the model information storage unit 122.

[0099] The learning unit 133 performs a learning process. The learning unit 133 performs various learning. The learning unit 133 learns various information based on the information acquired by the acquisition unit 131. The learning unit 133 learns (generates) a model. The learning unit 133 learns various information such as a model. The learning unit 133 generates a model by learning. The learning unit 133 learns the model using various machine learning techniques. For example, the learning unit 133 learns parameters of the model (network). The learning unit 133 learns the model using various machine learning techniques.

[0100] The learning unit 133 generates a model by machine learning using the learning data. The learning unit 133 receives data used as a search query as input and generates a model that outputs information indicating parameter values ​​corresponding to the input data. The learning unit 133 generates a model that is a classification model that outputs a classification result indicating parameter values ​​corresponding to the input data.

[0101] The learning unit 133 learns a model so that, when a query object included in the learning data is input, a classification result corresponding to a parameter value associated with the query object is output. The learning unit 133 generates a model that outputs a parameter value corresponding to the input data. The learning unit 133 learns a model so that, when a query object included in the learning data is input, a parameter value associated with the query object is output.

[0102] The learning unit 133 generates various learning models such as the model M1. The learning unit 133 learns network parameters. For example, the learning unit 133 learns network parameters of various learning models such as the model M1. The learning unit 133 performs a learning process using the learning data stored in the model information storage unit 122 to generate various learning models such as the model M1. For example, the learning unit 133 generates a model used to estimate the value of a parameter corresponding to a query. The learning unit 133 generates various learning models such as the model M1 by learning network parameters of various learning models such as the model M1.

[0103] The learning unit 133 performs a learning process based on the learning data (teacher data) stored in the model information storage unit 122. The learning unit 133 performs a learning process using the learning data stored in the model information storage unit 122, thereby generating various learning models such as model M1.

[0104] For example, the learning unit 133 performs a learning process by a method such as backpropagation so that the estimated values ​​of parameters output by the model M1 approach correct answer information (information indicating desired parameter values) associated with the input information (query) input to the model M1. For example, the learning unit 133 performs a learning process so that the estimated values ​​of parameters output by the model M1 to which query information has been input as input information approach the parameter values ​​(correct answer information) associated with the input query information.

[0105] As described above, the method of learning by the learning unit 133 is not particularly limited, but for example, learning data in which data (input information) and its correct answer information (output information) are linked may be prepared, and the learning data may be input to a computation model based on a multi-layer neural network for learning. In addition, a method based on DNN such as CNN or 3D-CNN may be used. Specifically, the learning method by the learning unit 133 may be a method based on U-net. When targeting time-series data such as voice, the learning unit 133 may use a method based on a recurrent neural network (RNN) or LSTM (Long Short-Term Memory units), which is an extension of RNN.

[0106] (Processing section 134) The processing unit 134 processes various information. The processing unit 134 performs information processing based on the information stored in the storage unit 120. The processing unit 134 functions as an inference unit that performs inference processing. The processing unit 134 performs inference processing by using a model stored in the model information storage unit 122. For example, the processing unit 134 performs inference processing to estimate a parameter value corresponding to the query by inputting a query to the model M1 and causing the model M1 to output a parameter value. For example, the processing unit 134 performs inference processing to estimate a parameter value "ε" corresponding to the query by using the model M1.

[0107] The processing unit 134 executes a search process using an inference result using the model. The processing unit 134 sets parameter values ​​based on the output of the model and executes the search process. The processing unit 134 inputs a target query to the model, sets parameter values ​​to values ​​corresponding to the output of the model, and executes the search process.

[0108] The processing unit 134 executes a search process using the parameter values ​​estimated by the inference process. The processing unit 134 functions as a search processing unit that searches for various information. For example, the processing unit 134 inputs a query (target query) to be used in the search process into the model M1 and estimates the parameter values ​​corresponding to the target query. Then, when executing a search process using the target query, the processing unit 134 sets the parameters to the estimated values. Then, the processing unit 134 executes a search process for the target query using the parameters set to the estimated values.

[0109] The processing unit 134 extracts similar objects (similar nodes) that are objects (nodes) similar to the target query. The processing unit 134 searches for objects by searching a graph (graph data). When the acquisition unit 131 acquires a target query, the processing unit 134 searches for objects similar to the target query by searching the graph.

[0110] The processing unit 134 extracts objects similar to the target query by searching the graph. The processing unit 134 extracts objects similar to the target query by searching the graph based on the processing procedure shown in Fig. 7. The processing unit 134 searches the graph using a range AR1 within a radius r centered on the target query and a range AR2 within a radius r(1+ε) centered on the target query, and extracts nearby nodes.

[0111] (Provider 135) The providing unit 135 provides various information. For example, the providing unit 135 provides various information to the terminal device 10 and the information providing device 50. The providing unit 135 transmits various information to the terminal device 10 and the information providing device 50. The providing unit 135 provides various information based on the various information acquired by the acquiring unit 131. The providing unit 135 provides various information based on the various information generated by the generating unit 132.

[0112] The providing unit 135 provides various information based on the various information learned by the learning unit 133. The providing unit 135 provides various information based on the model learned by the learning unit 133. The providing unit 135 provides various information based on the various information processed by the processing unit 134. For example, the providing unit 135 transmits information to the terminal device 10.

[0113] The providing unit 135 provides the object ID extracted (searched) by the processing unit 134 to the terminal device 10 and the information providing device 50. The providing unit 135 provides the object ID extracted by the processing unit 134 through a search to the information providing device 50. The providing unit 135 provides the object ID extracted by the processing unit 134 to the terminal device 10 and the information providing device 50 as information indicating a vector corresponding to the query.

[0114] The providing unit 135 provides information on similar objects extracted by the processing unit 134. The providing unit 135 provides information on similar objects to the terminal device 10 used by a predetermined user. The providing unit 135 provides search results to the sender of the query. For example, in one example, the providing unit 135 provides information indicating objects extracted as similar objects of the target query to the sender of the search request. The information processing device 100 transmits information indicating similar objects of the target query to the terminal device 10 used by the user who made the search request.

[0115] [4. Information processing flow] Next, the flow of each information process will be described with reference to Fig. 6 and Fig. 7. Fig. 6 and Fig. 7 are flowcharts showing an example of information processing according to the embodiment. Specifically, Fig. 6 is a flowchart showing an example of generation processing according to the embodiment. Fig. 7 is a flowchart showing an example of search processing according to the embodiment.

[0116] [4-1. Generation process flow] First, a process example shown in Fig. 6 will be described. As shown in Fig. 6, the information processing device 100 acquires an index for multiple objects to be the target of data search, a query object for a search query in the data search, and reference information indicating a first accuracy required when performing a search with multiple objects as search targets (step S101).

[0117] The information processing device 100 generates a correct answer search result indicating an object that is a correct answer in a search targeting a plurality of objects using the query object (step S102).

[0118] The information processing device 100 repeatedly executes an extraction process that uses an index to extract nearby objects from among multiple objects that correspond to a query object by varying the values ​​of search parameters, and generates learning data to be used for training a model that estimates the parameter values ​​by associating the query object with the parameter values ​​when the difference between the first accuracy and the second accuracy based on a comparison between the result of the extraction process and the correct search result satisfies a predetermined condition (step S103).

[0119] [4-2. Search process flow] Next, the flow of the search process will be described with reference to FIG. 7 as an example. FIG. 7 is a flowchart showing an example of information processing according to the embodiment. Specifically, FIG. 7 is a flowchart showing an example of search processing using graph data. Note that the search processing shown in FIG. 7 also includes a selection process. The search process described below is performed by the information processing device 100. Also, the object referred to below may be read as a node. Note that the search using graph data by the information processing device 100 is not limited to the following, and may be performed by various procedures.

[0120] Here, the neighborhood set N(G, y) is a set of nearby objects associated by edges attached to node y. For example, the neighborhood set N(G, y) is a set of objects (nodes) connected by outgoing edges from node y. "G" may be predetermined graph data (for example, graphs GR11, GR21, etc.). For example, the information processing device 100 executes a k-nearest neighbor search process.

[0121] For example, the information processing device 100 sets the radius r of the hypersphere to ∞ (infinity) (step S300) and extracts a set S from an existing object set (step S301). For example, the information processing device 100 may extract an object (node) determined (selected) as a starting node as the set S. Also, for example, the hypersphere is a virtual sphere indicating a search range. Note that the objects (nodes) included in the set S extracted in step S301 are also included in an initial set of the set R of search results (extraction candidates). Also, the objects (nodes) included in the set S extracted in step S301 may be included in a set C. The set C is provided for convenience in order to avoid duplicate searches, and may be set to an empty set at the start of processing.

[0122] Next, the information processing device 100 extracts an object having the shortest distance from the search query object y among the objects included in the set S, and sets the object as object s (step S302). For example, in the example of FIG. 1, the information processing device 100 extracts one node as object s (target node) from the set S including node N2, which is the starting point node corresponding to the query QE1, which is the object y. Next, the information processing device 100 excludes object s from the set S (step S303). For example, in the example of FIG. 1, the information processing device 100 excludes node N2, which is the starting point node, from the set S.

[0123] Next, the information processing device 100 judges whether the distance d(s, y) between the object s and the object y exceeds r(1+ε) (step S304). Here, ε is an extension factor, and r(1+ε) is a value indicating the radius of the search range (only nodes within this range are searched for. By making it larger than the search range, the accuracy can be improved). If the distance d(s, y) between the object s and the object y exceeds r(1+ε) (step S304: Yes), the information processing device 100 outputs the set R as a neighborhood set of the object y (step S305), and ends the process.

[0124] If the distance d(s, y) between object s and search query object y does not exceed r(1+ε) (step S304: No), the information processing device 100 selects one object that is not included in set C from among the objects that are elements of the neighborhood set N(G, s) of object s, based on a predetermined criterion, and stores the selected object u in set C (step S306). For example, in the example of Fig. 1, the information processing device 100 selects the node (e.g., node N43) that is closest to query QE1 as object u from nodes N1, N43, etc. that are connecting nodes of node N2.

[0125] Next, the information processing device 100 determines whether the distance d(u, y) between the object u and the object y is r(1+ε) or less (step S307). If the distance d(u, y) between the object u and the object y is r(1+ε) or less (step S307: Yes), the information processing device 100 adds the object u to the set S (step S308). If the distance d(u, y) between the object u and the object y is not r(1+ε) or less (step S307: No), the information processing device 100 performs the determination (processing) of step S309.

[0126] Next, the information processing device 100 determines whether the distance d(u, y) between the object u and the object y is equal to or less than r (step S309). If the distance d(u, y) between the object u and the object y exceeds r, the information processing device 100 performs the determination (processing) of step S315. If the distance d(u, y) between the object u and the object y is not equal to or less than r (step S309: No), the information processing device 100 performs the determination (processing) of step S315.

[0127] If the distance d(u, y) between objects u and y is less than or equal to r (step S309: Yes), the information processing device 100 adds object u to set R (step S310). Then, the information processing device 100 determines whether the number of objects included in set R exceeds ks (step S311). The predetermined number ks is a natural number that is determined arbitrarily. For example, ks indicates the number of objects extracted in the search, and may be any value such as "3", "20", or "100". If the number of objects included in set R does not exceed ks (step S311: No), the information processing device 100 performs the determination (processing) of step S313.

[0128] If the number of objects included in set R exceeds ks (step S311: Yes), information processing device 100 excludes from set R the object that is the longest (farthest) distance from object y among the objects included in set R (step S312).

[0129] Next, the information processing device 100 judges whether the number of objects included in the set R matches ks (step S313). If the number of objects included in the set R does not match ks (step S313: No), the information processing device 100 performs the judgment (processing) of step S315. If the number of objects included in the set R matches ks (step S313: Yes), the information processing device 100 sets the distance between the object y and the object that is the longest (farthest) distance from the object y among the objects included in the set R to a new r (step S314).

[0130] Then, the information processing device 100 judges whether or not a number of objects corresponding to a threshold value have been selected from the objects that are elements of the neighborhood set N(G,s) of the object s (step S315). For example, in the example of FIG. 1, the information processing device 100 selects the output value "e p If the number of objects corresponding to the threshold value has not been selected from the objects that are elements of the neighborhood set N(G,s) of object s (step S315: No), the information processing device 100 returns to step S306 and repeats the process.

[0131] When the number of objects corresponding to the threshold value has been selected from the objects that are elements of the neighborhood set N(G,s) of the object s (step S315: Yes), the information processing device 100 determines whether the set S is an empty set or not (step S316). Note that the information processing device 100 may perform the process of step S316 if all objects in the neighborhood set N(G,s) of the object s have been selected even before selecting objects from the neighborhood set N(G,s) of the object s up to the number corresponding to the threshold value. That is, when the number of objects in the neighborhood set N(G,s) of the object s is equal to or less than the threshold value and all objects in the neighborhood set N(G,s) have been selected, the information processing device 100 may perform the process of step S316 in the same manner as when step S315 is Yes.

[0132] If the set S is not an empty set (step S316: No), the information processing device 100 returns to step S302 and repeats the process. If the set S is an empty set (step S316: Yes), the information processing device 100 outputs the set R and ends the process (step S317). For example, in the example of FIG. 1, the information processing device 100 outputs nodes N2, N43, N1, etc. included in the set R as search results corresponding to the query QE1 (input object y).

[0133] When executing a search process for generating learning data, the information processing device 100 collects objects included in the set R as data for calculating accuracy. In addition, for example, when the information processing device 100 executes a search process in response to a search request from a user, the information processing device 100 may provide objects (nodes) included in the set R as search results corresponding to the search query (input object y) to the terminal device 10 or the like used by the user who made the search request.

[0134] 5. Effects As described above, the information processing device 100 according to the embodiment includes an acquisition unit 131, a first generation unit (the generation unit 132 in the embodiment), and a second generation unit (the generation unit 132 in the embodiment). The acquisition unit 131 acquires an index for a plurality of objects to be the target of a data search, a query object for a search query in a data search, and reference information indicating a first accuracy required when performing a search with a plurality of objects as search targets. The generation unit 132 repeatedly executes an extraction process for extracting a nearby object corresponding to the query object from among the plurality of objects using the index by varying the value of a parameter related to the search, and generates learning data used for learning a model that estimates the value of a parameter by associating the query object with the value of the parameter when the difference between the first accuracy and the second accuracy based on a comparison between the result of the extraction process and the correct search result satisfies a predetermined condition.

[0135] In this way, the information processing device 100 according to the embodiment can generate learning data for generating a model that estimates parameters when using an index by adjusting parameter values ​​so as to satisfy a desired accuracy. Therefore, the information processing device 100 can generate a model that estimates parameter values ​​of an index in response to a query.

[0136] Moreover, in the information processing device 100 according to the embodiment, the acquiring unit 131 acquires an index used in the approximate neighborhood search.

[0137] In this way, the information processing device 100 according to the embodiment can generate learning data for generating a model that estimates parameters when using an index used in an approximate neighborhood search by adjusting parameter values ​​to achieve a desired accuracy. Therefore, the information processing device 100 can generate a model that estimates parameter values ​​of an index in response to a query.

[0138] Moreover, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires, as an index, a graph index in which a plurality of objects are connected by edges.

[0139] In this way, the information processing device 100 according to the embodiment can generate learning data for generating a model that estimates parameters when using a graph index in which multiple objects are connected by edges, by adjusting parameter values ​​so as to satisfy a desired accuracy. Therefore, the information processing device 100 can generate a model that estimates index parameter values ​​in response to a query.

[0140] Furthermore, in the information processing device 100 according to the embodiment, the generation unit 132 repeatedly executes the extraction process by gradually changing the parameter value in the direction of change when the accuracy of the result of the extraction process is high, and generates learning data by associating the parameter value when the difference between the second accuracy and the first accuracy satisfies a predetermined condition with a query object.

[0141] In this way, the information processing device 100 according to the embodiment can generate learning data efficiently by repeatedly executing the extraction process while gradually changing the parameter value in the direction of change, which results in a high accuracy of the extraction process result. Therefore, the information processing device 100 can generate a model that estimates the parameter value of an index in response to a query.

[0142] Furthermore, in the information processing device 100 according to the embodiment, the generation unit 132 generates learning data by repeatedly executing the extraction process while gradually increasing the parameter value, and associating the parameter value when the difference between the second accuracy and the first accuracy satisfies a predetermined condition with a query object.

[0143] In this way, the information processing device 100 according to the embodiment can efficiently generate learning data by repeatedly executing the extraction process while gradually increasing the parameter value. Therefore, the information processing device 100 can generate a model that estimates the parameter value of an index in response to a query.

[0144] In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires a query object, which is an object selected from a group of objects used for learning, and an index targeting a plurality of objects excluding the object selected as the query object from the group of objects.

[0145] In this way, the information processing device 100 according to the embodiment can efficiently generate learning data by selecting a query from a group of objects used for learning. Therefore, the information processing device 100 can generate a model that estimates the value of an index parameter according to a query.

[0146] Moreover, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires a query object which is an object randomly selected from a group of objects.

[0147] In this way, the information processing device 100 according to the embodiment can efficiently generate learning data by randomly selecting a query from a group of objects. Therefore, the information processing device 100 can generate a model that estimates the value of an index parameter in response to a query.

[0148] Moreover, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires an index targeting a plurality of objects that are an object group used for learning, and a query object that is an object other than the object group.

[0149] In this way, the information processing device 100 according to the embodiment can generate high-quality learning data by selecting a query from outside the object group and by selecting appropriate information as a query. Therefore, the information processing device 100 can generate a model that estimates the value of an index parameter in response to a query.

[0150] Moreover, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires a query object from a query object group different from the object group.

[0151] In this way, the information processing device 100 according to the embodiment can generate high-quality learning data by selecting a query from a query object group different from an object group and selecting appropriate information as a query. Therefore, the information processing device 100 can generate a model that estimates the value of an index parameter according to a query.

[0152] Moreover, the information processing device 100 according to the embodiment includes a learning unit 133. The learning unit 133 generates a model by machine learning using learning data.

[0153] In this manner, the information processing device 100 according to the embodiment can generate a model that estimates the value of an index parameter in response to a query by generating a model through machine learning using learning data.

[0154] Moreover, in the information processing device 100 according to the embodiment, the learning unit 133 receives data used as a search query as input and generates a model that outputs information indicating parameter values ​​corresponding to the input data.

[0155] In this way, the information processing device 100 according to the embodiment can generate a model that takes data used as a search query as input and outputs information indicating parameter values ​​corresponding to the input data, thereby generating a model that estimates the parameter values ​​of an index in response to a query.

[0156] Moreover, in the information processing device 100 according to the embodiment, the learning unit 133 generates a model that is a classification model that outputs a classification result indicating the value of a parameter corresponding to input data.

[0157] In this way, the information processing device 100 according to the embodiment can generate a model that estimates the value of an index parameter in response to a query by generating a model that is a classification model that outputs a classification result indicating the value of a parameter corresponding to input data.

[0158] Moreover, in the information processing device 100 according to the embodiment, the learning unit 133 learns a model so that when a query object included in the learning data is input, a classification result corresponding to the value of a parameter associated with the query object is output.

[0159] In this way, the information processing device 100 according to the embodiment can generate a model that estimates the value of an index parameter in response to a query by training a model so that, when a query object included in the learning data is input, a classification result corresponding to the value of a parameter associated with the query object is output.

[0160] Moreover, in the information processing device 100 according to the embodiment, the learning unit 133 generates a model that outputs parameter values ​​corresponding to input data.

[0161] In this way, the information processing device 100 according to the embodiment can generate a model that estimates the value of an index parameter in response to a query by generating a model that outputs the value of a parameter corresponding to input data.

[0162] In addition, in the information processing device 100 according to the embodiment, the learning unit 133 learns a model so that, when a query object included in the learning data is input, the model outputs a parameter value associated with the query object.

[0163] In this way, the information processing device 100 according to the embodiment can generate a model that estimates the value of an index parameter in response to a query by training a model to output the value of a parameter associated with a query object included in the training data when the query object is input.

[0164] Moreover, the information processing device 100 according to the embodiment includes a search processing unit (in the embodiment, a processing unit 134). The processing unit 134 executes a search process using an inference result using a model.

[0165] In this way, the information processing device 100 according to the embodiment can perform search processing using the inference results using the model, thereby performing search processing using the parameter values ​​of the index estimated in response to the query.

[0166] Moreover, in the information processing device 100 according to the embodiment, the processing unit 134 sets parameter values ​​based on the output of the model and executes search processing.

[0167] In this way, the information processing device 100 according to the embodiment can perform search processing using the parameter values ​​of the index estimated in response to the query by setting parameter values ​​based on the output of the model and performing search processing.

[0168] In the information processing device 100 according to the embodiment, the acquisition unit 131 acquires a target query that is a search query to be processed. The processing unit 134 inputs the target query to a model, sets parameter values ​​to values ​​corresponding to the output of the model, and executes search processing.

[0169] In this way, the information processing device 100 according to the embodiment can execute a search process using the parameter values ​​of the index estimated according to the query by inputting the target query into the model, setting the parameter values ​​to values ​​corresponding to the output of the model, and executing a search process.

[0170] [6. Hardware Configuration] Moreover, the information processing device 100 and the terminal device 10 according to the above-described embodiment are realized by a computer 1000 having a configuration as shown in Fig. 8, for example. The information processing device 100 will be described below as an example. Fig. 8 is a diagram showing an example of a hardware configuration. The computer 1000 is connected to an output device 1010 and an input device 1020, and has a configuration in which a calculation device 1030, a primary storage device 1040, a secondary storage device 1050, an output I / F (Interface) 1060, an input I / F 1070, and a network I / F 1080 are connected by a bus 1090.

[0171] The arithmetic device 1030 operates based on programs stored in the primary storage device 1040 and the secondary storage device 1050, programs read from the input device 1020, and the like, and executes various processes. The arithmetic device 1030 is realized by, for example, a CPU (Central Processing Unit), an MPU (Micro Processing Unit), an ASIC (Application Specific Integrated Circuit), an FPGA (Field Programmable Gate Array), or the like.

[0172] The primary storage device 1040 is a memory device such as a RAM (Random Access Memory) that primarily stores data used by the arithmetic device 1030 for various calculations. The secondary storage device 1050 is a storage device in which data used by the arithmetic device 1030 for various calculations and various databases are registered, and is realized by a ROM (Read Only Memory), an HDD (Hard Disk Drive), an SSD (Solid State Drive), a flash memory, or the like. The secondary storage device 1050 may be an internal storage device or an external storage device. The secondary storage device 1050 may be a removable storage medium such as a USB memory or a Secure Digital (SD) memory card. The secondary storage device 1050 may be a cloud storage device (online storage device), a NAS (Network Attached Storage), a file server, or the like.

[0173] The output I / F 1060 is an interface for transmitting information to be output to an output device 1010 that outputs various types of information, such as a display, a projector, a printer, etc., and is realized by a connector conforming to a standard such as USB (Universal Serial Bus), DVI (Digital Visual Interface), or HDMI (registered trademark) (High Definition Multimedia Interface). The input I / F 1070 is an interface for receiving information from various input devices 1020, such as a mouse, a keyboard, a keypad, a button, a scanner, etc., and is realized by a USB, etc.

[0174] Furthermore, the output I / F 1060 and the input I / F 1070 may be wirelessly connected to the output device 1010 and the input device 1020, respectively. That is, the output device 1010 and the input device 1020 may be wireless devices.

[0175] The output device 1010 and the input device 1020 may be integrated into one device, such as a touch panel. In this case, the output I / F 1060 and the input I / F 1070 may also be integrated into one device, such as an input / output I / F.

[0176] The input device 1020 may be a device that reads information from, for example, an optical recording medium such as a CD (Compact Disc), a DVD (Digital Versatile Disc), or a PD (Phase change rewritable Disk), a magneto-optical recording medium such as an MO (Magneto-Optical disk), a tape medium, a magnetic recording medium, or a semiconductor memory.

[0177] The network I / F 1080 receives data from other devices via the network N and sends it to the arithmetic device 1030, and also transmits data generated by the arithmetic device 1030 to other devices via the network N.

[0178] The arithmetic unit 1030 controls the output device 1010 and the input device 1020 via the output I / F 1060 and the input I / F 1070. For example, the arithmetic unit 1030 loads a program from the input device 1020 or the secondary storage device 1050 onto the primary storage device 1040 and executes the loaded program.

[0179] For example, when the computer 1000 functions as the information processing device 100, the arithmetic unit 1030 of the computer 1000 executes a program loaded onto the primary storage device 1040 to realize the functions of the control unit 130. The arithmetic unit 1030 of the computer 1000 may also load a program acquired from another device via the network I / F 1080 onto the primary storage device 1040 and execute the loaded program. The arithmetic unit 1030 of the computer 1000 may also cooperate with the other device via the network I / F 1080 and use functions, data, etc. of a program by calling them from another program of the other device.

[0180] [7. Other] Furthermore, among the processes described in the above embodiments and modifications, all or part of the processes described as being performed automatically can be performed manually, or all or part of the processes described as being performed manually can be performed automatically by a known method. In addition, the information including the processing procedures, specific names, various data and parameters shown in the above documents and drawings can be changed arbitrarily unless otherwise specified. For example, the various information shown in each drawing is not limited to the illustrated information.

[0181] In addition, each component of each device shown in the figure is a functional concept, and does not necessarily have to be physically configured as shown in the figure. In other words, the specific form of distribution and integration of each device is not limited to that shown in the figure, and all or part of them can be functionally or physically distributed and integrated in any unit according to various loads, usage conditions, etc.

[0182] Furthermore, the above-described embodiments and modifications can be appropriately combined as long as the processing contents are not contradictory.

[0183] Furthermore, the above-mentioned "section, module, unit" can be read as "means" or "circuit", etc. For example, an acquisition section can be read as an acquisition means or an acquisition circuit. [Explanation of symbols]

[0184] 1. Information Processing Systems 100 Information processing device 121 Learning data storage unit 122 Model information storage unit 130 Control section 131 Acquisition Department 132 Generation part 133 Learning Department 134 Processing section (search processing section) 135 Provision Department 10 Terminal Equipment 50 Information provision device N Network

Claims

1. an acquisition unit that acquires an index for a plurality of objects to be the target of a data search, a query object for a search query in the data search, and reference information indicating a first accuracy required when performing a search using the plurality of objects as the search targets; a first generation unit that generates a correct answer search result indicating an object that is a correct answer in a search targeting the plurality of objects by using the query object; a second generation unit that generates learning data to be used for learning a model that estimates the value of the parameter by repeatedly executing an extraction process, in which an object near the query object is extracted from among the plurality of objects using the index, by varying a value of a parameter related to a search, and associating the value of the parameter with the query object when a difference between a second accuracy based on a comparison between a result of the extraction process and the correct search result and the first accuracy satisfies a predetermined condition; An information processing device comprising:

2. The acquisition unit is Obtain the index used for approximate neighborhood search 2. The information processing apparatus according to claim 1,

3. The acquisition unit is A graph index in which the plurality of objects are connected by edges is obtained as the index.

3. The information processing apparatus according to claim 2.

4. The second generation unit is The extraction process is repeatedly performed by gradually changing the value of the parameter in a direction of change in which the accuracy of the result of the extraction process is high, and the parameter value when the difference between the second accuracy and the first accuracy satisfies a predetermined condition is associated with the query object, thereby generating the learning data.

2. The information processing apparatus according to claim 1,

5. The second generation unit is The parameter value is increased stepwise to repeatedly execute the extraction process, and the parameter value at the time when the difference between the second accuracy and the first accuracy satisfies a predetermined condition is associated with the query object, thereby generating the learning data.

2. The information processing apparatus according to claim 1,

6. The acquisition unit is The query object is an object selected from a group of objects used for learning, and the index is obtained for the plurality of objects excluding the object selected as the query object from the group of objects.

2. The information processing apparatus according to claim 1,

7. The acquisition unit is Obtain the query object, which is an object randomly selected from the group of objects.

7. The information processing apparatus according to claim 6,

8. The acquisition unit is The index is obtained for the plurality of objects that are a group of objects used for learning, and the query object is obtained for the object other than the group of objects.

2. The information processing apparatus according to claim 1,

9. The acquisition unit is Obtain the query object from a query object group different from the object group.

9. The information processing apparatus according to claim 8,

10. a learning unit that generates the model by machine learning using the learning data; 2. The information processing apparatus according to claim 1,

11. The learning unit is The model is generated by inputting data used as a search query and outputting information indicating values ​​of the parameters corresponding to the input data.

11. The information processing apparatus according to claim 10,

12. The learning unit is Generate the model, which is a classification model that outputs a classification result indicating the value of the parameter corresponding to the input data.

12. The information processing apparatus according to claim 11 .

13. The learning unit is The model is trained so that, when a query object included in the learning data is input, a classification result corresponding to the value of the parameter associated with the query object is output.

13. The information processing apparatus according to claim 12.

14. The learning unit is Generate the model that outputs values ​​of the parameters corresponding to the input data.

12. The information processing apparatus according to claim 11 .

15. The learning unit is The model is trained so that, when a query object included in the learning data is input, a value of the parameter associated with the query object is output.

15. The information processing apparatus according to claim 14,

16. a search processing unit that executes a search process using an inference result obtained by using the model; 2. The information processing apparatus according to claim 1,

17. The search processing unit: Setting values ​​of the parameters based on the output of the model and performing the search process.

17. The information processing apparatus according to claim 16,

18. The acquisition unit is Obtain a target query, which is a search query to be processed; The search processing unit: inputting the target query into the model, setting the values ​​of the parameters to values ​​corresponding to the outputs of the model, and executing the search process; 17. The information processing apparatus according to claim 16,

19. 1. A computer-implemented information processing method, comprising: an acquisition step of acquiring an index for a plurality of objects to be the target of a data search, a query object for a search query in the data search, and reference information indicating a first accuracy required when performing a search using the plurality of objects as the search targets; a first generation step of generating a correct answer search result indicating an object that is a correct answer in a search targeting the plurality of objects using the query object; a second generation step of repeatedly executing an extraction process using the index to extract nearby objects corresponding to the query object from among the plurality of objects while varying a value of a search parameter, and generating learning data to be used for learning a model that estimates the value of the parameter by associating the query object with a value of the parameter when a difference between a second accuracy based on a comparison between a result of the extraction process and the correct search result and the first accuracy satisfies a predetermined condition; 13. An information processing method comprising:

20. an acquisition step of acquiring an index for a plurality of objects to be the target of a data search, a query object for a search query in the data search, and reference information indicating a first accuracy required when performing a search using the plurality of objects as the search targets; a first generation step of generating a correct answer search result indicating an object that is a correct answer in a search targeting the plurality of objects using the query object; a second generation step of repeatedly executing an extraction process using the index to extract nearby objects corresponding to the query object from among the plurality of objects while varying a value of a search parameter, and generating learning data to be used for learning a model that estimates the value of the parameter by associating the query object with a value of the parameter when a difference between a second accuracy based on a comparison between a result of the extraction process and the correct search result and the first accuracy satisfies a predetermined condition; An information processing program characterized by causing a computer to execute the above.

Citation Information

Patent Citations

  • Information retrieval system

    JP2020009140A

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

    JP2022035382A

  • Conductor for connection of semiconductor device

    JP1987093335A

  • Power source plug slip alarm

    JP1988000982A

  • Ultrasonic transformer

    JP1988011000A