Information Processing Apparatus, Information Processing Method, and Program

Abstract

To perform accurate analysis on data in every category.SOLUTION: An information processing apparatus has: identification means that identifies a category of input data with a category identification model; and control means that, on the basis of the degree of adaption of data belonging to the N-th category to analysis models for the first to the M-th categories for performing analysis on data in the first to the M-th categories, selects, from analysis models for the first to the M-th categories, an analysis model to be applied to the input data identified to be in the N-th category by the identification means.SELECTED DRAWING: Figure 6

Description

Technical Field

[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.

Background Art

[0002] In recent years, technologies for extracting and analyzing useful information from various types of data such as images, texts, and time-series data have been proposed, and in particular, by using machine learning technologies, it has become possible to perform highly accurate analysis. Patent Document 1 describes a system that uses a machine learning model to identify to which category unknown data belongs and executes an analysis task suitable for the data of the identified category (hereinafter, specific category).

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Non-Patent Documents

[0004]

Non-Patent Document 1

Non-Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0005] When executing an analysis task suitable for each specific category as in Patent Document 1, it is necessary to prepare an analysis model for executing the analysis task for each specific category. At this time, if the types of specific categories are to be increased, a new analysis model is required. However, for example, due to hardware resource constraints, it may be difficult to newly generate an analysis model.

[0006] An object of the present invention is to perform highly accurate analysis on data for each category.

Means for Solving the Problems

[0007] The present invention is Learning means for learning an identification model that identifies data categories, wherein the learning means uses teacher data labeled with a second category, which is a category of an analysis model for which the fitness of data belonging to a first category with respect to an analysis model for each category for analyzing data of each category is equal to or greater than a threshold value, for learning the identification model. characterized by the following.

Effects of the Invention

[0008] According to the present invention, highly accurate analysis can be performed on data for each category.

Brief Description of the Drawings

[0009]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Mode for Carrying Out the Invention

[0010] Hereinafter, embodiments will be described with reference to the accompanying drawings.

[0011] 〔Embodiment 1〕 In this embodiment, a system for detecting a specific part (for example, a face) of an identified category by applying a detection model that identifies the category of an object in an input image and detects the specific part of the identified category from the image will be described. The process of detecting a specific part of a category from an image is an example of an analysis process. Note that the input data is not limited to images, and is not particularly limited, such as documents and time-series data.

[0012] FIG. 1 shows an example of the hardware configuration of an information processing apparatus according to this embodiment. The information processing apparatus 1 includes a CPU 11, a ROM 12, a RAM 13, a secondary storage device 14, an input device 15, and a display device 16. These components are interconnected via a connection bus 17. The CPU (Central Processing Unit) 11 controls the entire information processing apparatus 1. The CPU 11 executes the control programs stored in the ROM 12 and the like, thereby executing the processes of each flowchart described later. Note that, instead of or together with the CPU, a GPU (Graphics Processing Unit) may be used.

[0013] The ROM 12 is a non-volatile memory that stores control programs and various parameter data. The RAM 13 is a volatile memory that temporarily stores images, control programs, and their execution results. The secondary storage device 14 is a rewritable secondary storage device such as a hard disk or a flash memory, and stores various data used in the flowcharts described later. For example, it stores input data, a dataset for learning, and processing results. These pieces of information are output to the RAM 13 and used by the CPU 11 to execute the control program. The input device 15 is a keyboard, a mouse, a touch panel device, etc., and inputs various user instructions. The display device 16 is a monitor and displays processing results, images, etc.

[0014] In this embodiment, the processing described later will be realized by software using the CPU 11, but a part or all of the processing described later may be realized by hardware. As the hardware, a dedicated circuit (ASIC), a processor (reconfigurable processor, DSP), etc. can be used. Further, the information processing device 1 has a communication unit for communicating with an external device, and may acquire input data, a control program, a dataset for learning, etc. from the external device via the communication unit, and may output processing results, etc. to the external device via the communication unit.

[0015] FIG. 2 shows a functional configuration example of the information processing device according to this embodiment. The information processing device 1 is configured to have functions as each functional unit shown in FIG. 2 by the CPU 11 executing a control program stored in the ROM 12 or the like. The information processing device 1 has an input data acquisition unit 202, a category identification unit 203, and an analysis unit 206.

[0016] The input data acquisition unit 202 acquires the image 201 as input data from an external device or the secondary storage device 14. The category identification unit 203 identifies the category of the image. In the present embodiment, numbers are assigned to the categories in serial order from the first to the Nth, and N categories (the first category, the second category, ···, the Mth category, and the Nth category (N > M)) are identified. The category identification unit 203 includes a category identification model for identifying the category of an object in the image.

[0017] The analysis unit 206 has a detection model for detecting the region of a specific part of the category (hereinafter referred to as the specific region) from the image. In the present embodiment, there are M detection models (the detection model 207 for the first category, the detection model 208 for the second category, ···, the detection model 209 for the Mth category) for each of the M categories. On the other hand, the analysis unit 206 does not have a detection model for the Nth category (N > M).

[0018] Each detection model is a machine learning model learned using a teacher dataset belonging to each category. The detection model 207 for the first category is a machine learning model learned using a teacher dataset belonging to the first category. Similarly, for the detection models for the second to Mth categories, they are machine learning models learned using teacher datasets belonging to their respective categories. The machine learning model is, for example, a model learned based on correct information of the specific region of the category using a multi-layer neural network. Known methods such as Non-Patent Document 1 are applied to the learning method.

[0019] <Fitness calculation process for the Nth category> The information processing apparatus 1 has a fitness calculation unit 214. Hereinafter, the process executed by the fitness calculation unit 214 will be described. In the present embodiment, the fitness calculation unit 214 uses the detection models for the first to Mth categories of the analysis unit 206 and the fitness calculation dataset 217 for the Nth category to generate a fitness calculation result 205. The generated fitness calculation result 205 is stored in the secondary storage device 14 or the like.

[0020] FIG. 3 is a flowchart showing the fitness calculation process for the N-th category. The fitness of the data of the N-th category with respect to the detection model for the m-th category is an index indicating whether the detection model for the m-th category can be applied to the detection of a specific region of the N-th category. The fitness of the data of the N-th category with respect to the detection model for the m-th category is represented by fitness A(N:m). In the following description, the notation of the steps (steps) is omitted by attaching S at the beginning for each step (step).

[0021] In S301, the fitness calculation unit 214 acquires the fitness calculation dataset 217 for the N-th category. The fitness calculation dataset 217 for the N-th category includes one or more sets of data consisting of an image in which an object of the N-th category is imaged and information (correct answer information) indicating a specific region within the image. The fitness calculation dataset 217 for the N-th category may be stored in the secondary storage device 14 or the like. The fitness calculation unit 214 may acquire the fitness calculation dataset 217 for the N-th category from the secondary storage device 14, or may acquire it from an external device.

[0022] In S302, the fitness calculation unit 214 stores 1 in m representing the category number. In S303, the fitness calculation unit 214 determines whether the detection model for the m-th category exists in the analysis unit 206. When the fitness calculation unit 214 determines that the detection model for the m-th category exists (Yes in S303), it acquires the detection model for the m-th category from the analysis unit 206 and proceeds to S304. When it determines that it does not exist (No in S303), it proceeds to S307.

[0023] In S304, the fitness calculation unit 214 applies the detection model for the m-th category to the fitness calculation dataset 217 for the N-th category acquired in S301. Specifically, the image of the fitness calculation dataset 217 for the N-th category is input to the detection model for the m-th category, and a region is detected from the image.

[0024] In S305, the fitness calculation unit 214 calculates the fitness A(N:m) based on the comparison result between the detection region detected by the detection model for the m-th category in S304 and the specific region represented by the ground truth information of the dataset 217 for calculating the fitness of the N-th category. Here, the difference between the regions is obtained as the comparison result. The difference between the regions is quantitatively given, for example, by the difference in the positions of the regions, the difference in the size / area, or a combination thereof. When the difference between the regions is large, it means that the detection model for the m-th category cannot detect the specific region of the N-th category, so it can be determined that the fitness A(N:m) is low. On the other hand, when the difference between the regions is small, it means that the detection model for the m-th category can detect the specific region of the N-th category, so it can be determined that the fitness A(N:m) is high. Note that the fitness A(N:m) may be represented by the reciprocal of the difference between the detection region of the detection model for the m-th category and the specific region of the N-th category, or may be represented by the size of the IoU (Intersection over Union) between the detection region and the specific region.

[0025] In S306, the fitness calculation unit 214 increments m representing the category number and proceeds to S303. By repeating the processes of S303 to S306, the fitness calculation unit 214 calculates the fitness A(N:m) for each of the M detection models including the detection model 207 for the first category to the detection model 209 for the M-th category. Finally, in S307, the fitness calculation unit 214 stores the M fitnesses A(N:m) (m = 1, 2, ···, M) in the secondary storage device 14 or the like as the fitness calculation result 205. Then, the processing of the series of flowcharts ends.

[0026] <Learning Process of Category Identification Model> The information processing apparatus 1 includes a teacher dataset generation unit 215 and a learning unit 213. Hereinafter, the processes executed by the teacher dataset generation unit 215 and the learning unit 213 will be described. The learning unit 213 learns a category discrimination model used in the category discrimination unit 203. The teacher dataset for learning the category discrimination model is composed of the teacher datasets 211 for the first category to the M-th category and the teacher dataset 212 for the N-th category. The teacher dataset 211 is stored in the secondary storage device 14 or the like. The teacher dataset 212 is generated by the teacher dataset generation unit 215. The teacher datasets 211 and 212 include a plurality of datasets each consisting of an image in which an object is imaged and label information indicating to which category the object in the image belongs.

[0027] In this embodiment, the case where only the fitness A(N:1) is high among the M fitnesses A(N:m) will be described. In this case, the teacher dataset 212 for the N-th category may be labeled with the sub-category of the first category and used for learning the category discrimination model. Alternatively, the teacher dataset 212 for the N-th category may be labeled with the N-th category as the (M + 1)-th category and used for learning the category discrimination model. If it is the former, the learning unit 213 learns to discriminate M categories from the first category to the M-th category, and if it is the latter, the learning unit 213 learns to discriminate (M + 1) categories with the N-th category added.

[0028] Hereinafter, the process executed by the teacher dataset generation unit 215 in the former case described above will be explained. In this embodiment, the teacher dataset generation unit 215 generates the teacher dataset 212 for the N-th category from the dataset 216 for the N-th category based on the fitness calculation result 205.

[0029] FIG. 4 is a flowchart showing the teacher data generation process for the N-th category. In S401, the teacher dataset generation unit 215 acquires the dataset 216 of the Nth category. The dataset 216 of the Nth category may be stored in the secondary storage device 14 or the like. The teacher dataset generation unit 215 may acquire the dataset 216 of the Nth category from the secondary storage device 14 or may acquire it from an external device. In S402, the teacher dataset generation unit 215 acquires the fitness A(N:m) of the data of the Nth category with respect to the detection model for the mth category from the fitness calculation result 205. As described above, assuming that only the fitness A(N:1) is high among the M fitnesses A(N:m), it will be described that only the fitness A(N:1) is acquired in S402. In S403, the teacher dataset generation unit 215 determines whether the fitness A(N:1) is equal to or greater than a predetermined threshold. If the teacher dataset generation unit 215 determines that the fitness A(N:1) is equal to or greater than the threshold (Yes in S403), it proceeds to S404. If it determines that the fitness A(N:1) is less than the threshold (No in S403), it proceeds to S405.

[0030] In S404, the teacher dataset generation unit 215 labels the dataset 216 of the Nth category with the first category as a subcategory and generates it as the teacher dataset 212 of the Nth category. Thereby, it becomes possible to advance the learning of the category discrimination model so that the Nth category approaches the first category. The generated teacher dataset 212 is stored in the secondary storage device 14 or the like. Then, the processing of this flowchart ends. In S405, the teacher dataset generation unit 215 labels the dataset 216 of the Nth category with a category different from the first category and generates it as the teacher dataset 212 of the Nth category. Thereby, it becomes possible to advance the learning of the category discrimination model so that the Nth category moves away from the first category. The generated teacher dataset 212 is stored in the secondary storage device 14 or the like. Then, the processing of this flowchart ends.

[0031] FIG. 5 is a flowchart showing the learning process of the category identification model. The learning unit 213 learns the category identification model using the teacher dataset 211 of the first category to the M-th category prepared in advance and the teacher dataset 212 of the N-th category generated by the teacher dataset generation unit 215.

[0032] In S501, the learning unit 213 acquires the teacher dataset 211 of the first category to the M-th category. In S502, the learning unit 213 acquires the teacher dataset 212 of the N-th category.

[0033] In S503, the learning unit 213 acquires the category identification model used in the category identification unit 203. In the present embodiment, the category identification model is a machine learning model for identifying the category of an object in an image. The machine learning model is a multi-layer neural network model. However, it is not limited to the multi-layer neural network model, and known machine learning models such as random forest and AdaBoost may be used. In S504, the learning unit 213 acquires mini-batch images as input images from the teacher datasets acquired in S501 and S502.

[0034] In S505, the learning unit 213 performs inference processing on the images acquired in S504 using the category identification model acquired in S503. In S506, the learning unit 213 acquires the label information corresponding to the images acquired in S504. In S507, the learning unit 213 calculates the value of the loss (loss function). In S508, the learning unit 213 applies the error backpropagation method to the loss calculated in S507 to calculate the gradient and obtains the update amount of the weights of the category identification model.

[0035] In S509, the learning unit 213 updates the weights of the category identification model. Specifically, a known learning method for a multi-layer neural network may be applied, and detailed description is omitted. In S510, the learning unit 213 outputs the category discrimination model with updated weights to the category discrimination unit 203. The learning unit 213 determines the weights of the category discrimination model by repeatedly executing the processes of S501 to S510 until the value of the loss and the discrimination accuracy converge. Then, the processing of a series of flowcharts is terminated.

[0036] <Detection model application process> FIG. 6 is a flowchart showing the detection model application process. The detection model application process is a process of identifying the category of input data of an unknown category, selecting a detection model corresponding to the identified category, and applying it to the input data. The information processing apparatus 1 includes an analysis application unit 704. Hereinafter, the process executed by the analysis application unit 704 will be described with reference to the flowchart of FIG. 6. Also, it is assumed that the category discrimination unit 203 uses a category discrimination model capable of discriminating the first to Nth categories. Also, it is assumed that the analysis unit 206 has detection models for the first to M categories. Also, it is assumed that the fitness calculation result 205 has M fitnesses A(N:m) (m = 1, 2, ···, M) for each of the detection models for the first to M categories of the data of the Nth category.

[0037] In S601, the input data acquisition unit 202 acquires an image 201 of an unknown category. In S602, the analysis application unit 204 acquires the fitness calculation result 205. In S603, the category discrimination unit 203 performs category discrimination on the image 201 using the category discrimination model and outputs the discrimination result.

[0038] In S604, the analysis application unit 204 determines whether the discrimination result is one of the first to M categories. If the analysis application unit 204 determines that it is one of the first to M categories (Yes in S604), the process proceeds to S605. If it determines that it is a category other than the first to M categories (No in S604), the process proceeds to S606. In S605, the analysis application unit 204 selects the detection model for the category output as the identification result. The subsequent process proceeds to S608. In this way, when the identification result represents the first to M categories (specific categories), the analysis application unit 204 selects the detection model for that category.

[0039] In S606, the analysis application unit 204 determines whether the identification result is the Nth category. If the analysis application unit 204 determines that it is not the Nth category (No in S606), the series of processes in this flowchart ends. If it determines that it is the Nth category (Yes in S606), the process proceeds to S607. In S607, the analysis application unit 204 searches for the category n with the highest degree of fitness A(N:n) (1 ≦ n ≦ M) for the Nth category from the fitness calculation result 205 obtained in S602, and selects the detection model for the nth category. In this way, when the identification result represents the Nth category (category other than the specific category), the analysis application unit 204 selects the detection model for the category from among the detection models for the first to M categories based on the degree of fitness with the Nth category. In S608, the analysis unit 206 applies the detection model selected in S605 or S607 to the image 201 obtained in S601, and outputs the detection result 210. Then the processing of this flowchart ends.

[0040] As in this embodiment, when the degree of fitness A(N:1) is high and the sub-category of the first category is labeled in the teacher dataset 212 of the Nth category for learning the category identification model, the learning progresses such that the Nth category approaches the first category. Therefore, in S603 of FIG. 6, the Nth category is identified as the first category, and the detection model for the first category is selected in S605. On the other hand, when the teacher dataset 212 of the Nth category is labeled with the Nth category and the category identification model is trained, the training proceeds such that the Nth category is identified as the Nth category. Therefore, in S603 of FIG. 6, the Nth category is identified as the Nth category, and in S607, the detection model is selected from the detection models for the first to Mth categories based on the fitness. In this embodiment, since the fitness A(N:1) is high, the detection model for the first category is selected.

[0041] In this embodiment, when only the detection model for a specific category exists and an image of a category other than the specific category is obtained, by selecting the detection model for the specific category according to the fitness, it becomes possible to perform a high-quality detection task. Also, by using the image of the category other than the specific category as teacher data for training the category identification model to identify the specific category with a high degree of fitness for that category, it becomes easier to identify the category suitable for applying the detection task. That is, by effectively utilizing the existing analysis model, it becomes possible to perform a high-precision analysis task on the data of the category other than the specific category.

[0042] 〔Embodiment 2〕 In Embodiment 1, a configuration was described in which a detection model that identifies the category of an object in an input image and detects a specific part of the identified category is applied. In this embodiment, a configuration will be described in which a summarization model that classifies the category of an input document (for example, business, science and technology, entertainment) such as news and summarizes the classified document is applied. The process of summarizing the document of the category is an example of an analysis process. Note that the document summarization task is assumed to perform abstractive summarization. Abstractive summarization is considered a difficult task to generate a high-quality summary for document generation compared to extractive summarization, and as a method to improve the quality, it is conceivable to use a machine learning model trained specifically for the category. Hereinafter, parts common to Embodiment 1 will be omitted from the description, and the description will focus on the differences from Embodiment 1.

[0043] Figure 7 shows an example of the functional configuration of the information processing apparatus according to the present embodiment. In the functional configuration of Figure 7, compared with the functional configuration of Figure 2, the input data acquisition unit 202 acquires the document 701 as input data, and the analysis unit 706 has a summary model specialized for each category. Also, the analysis application unit 704 selects the summary model, and the summary result 710 is output as the result of applying the summary model. In the present embodiment, the analysis unit 706 has M summary models (summary model 707 for the first category, summary model 708 for the second category, ···, summary model 709 for the Mth category) for each of the M categories. On the other hand, the analysis unit 706 does not have a summary model for the Nth category (N > M).

[0044] The category identification unit 203 identifies the category of the document. The category identification model of the category identification unit 203 is, for example, a machine learning model fine-tuned by a known category identification learning method based on the model of non-patent literature. Each summary model of the analysis unit 706 is, for example, a machine learning model fine-tuned by a known abstractive summarization learning method based on the model of non-patent literature 2.

[0045] For the fitness A(N:m), for example, the ROUGE score, which is a type of evaluation index for document summarization, is used. Although there are several variations of ROUGE, for example, in the case of ROUGE-L, the document summarized by the summary model is compared with the summary document represented by the correct information of the input document, and the score calculated based on the maximum sequence (number of words) that matches in both documents is used. The higher the ROUGE score, the higher the fitness is considered to be.

[0046] In this embodiment, among the M degrees of fitness A(N:m), the case where the degree of fitness A(N:1) for the first category summary model 707 of the Nth category and the degree of fitness A(N:2) for the second category summary model 708 of the Nth category are equal to or greater than a threshold value will be described. In this case, for the document of the Nth category, either the summary model for the first category or the summary model for the second category can be applied. On the other hand, it is assumed that the degrees of fitness A(N:3) to A(N:M) for the summary models for the third to Mth categories of the data of the Nth category are less than the threshold value, and none of the summary models for the third to Mth categories can be applied to the document of the Nth category. Hereinafter, the processing executed by the teacher data set generation unit 215 in this embodiment will be described.

[0047] FIG. 8 is a flowchart showing the teacher data generation process for the Nth category. In S801, the teacher data set generation unit 215 acquires the data set 216 of the Nth category. In S802, the teacher data set generation unit 215 acquires the degrees of fitness A(N:m) (m = 1, 2, ···, M) for each of the summary models for the first to Mth categories of the data of the Nth category from the fitness calculation result 205.

[0048] In S803, the teacher data set generation unit 215 stores 1 in m representing the category number. In S804, the teacher data set generation unit 215 determines whether or not the degree of fitness A(N:m) is equal to or greater than a predetermined threshold value. If the teacher data set generation unit 215 determines that the degree of fitness A(N:m) is equal to or greater than the threshold value (Yes in S804), it proceeds to S805. If it determines that the degree of fitness A(N:1) is less than the threshold value (No in S804), it proceeds to S806.

[0049] In S805, the teacher data set generation unit 215 labels the data set 216 of the Nth category with the mth category as a subcategory, and generates it as the teacher data set 212 of the Nth category. The generated teacher data set 212 is stored in the secondary storage device 14 or the like. The subsequent processing proceeds to S807. In S806, the teacher dataset generation unit 215 labels the dataset 216 of the Nth category with a category different from the mth category, and generates it as the teacher dataset 212 of the Nth category. The generated teacher dataset 212 is stored in the secondary storage device 14 or the like. The subsequent process proceeds to S807.

[0050] In S807, the teacher dataset generation unit 215 increments m representing the category number and proceeds to S808. In S808, if m ≤ M, the teacher dataset generation unit 215 proceeds to S804, and if m > M, the series of processes of this flowchart is terminated. In this embodiment, the data of the Nth category is treated as the teacher data of both the sub - categories of the first category and the sub - categories of the second category.

[0051] In this embodiment, the Nth category is treated as the sub - categories of the first category and the second category. However, when there are two or more summary models with a fitness A(N:m) greater than or equal to the threshold value, the category of any one of the summary models may be selected as the sub - category. For example, the categories of the top n summary models with a high fitness A(N:m) may be selected as the sub - categories. In this case, if n = 1, the category of the summary model with the highest fitness is used as the sub - category.

[0052] Subsequently, with reference to FIG. 5, the learning process of the category identification model of this embodiment will be described. Here, the differences from Embodiment 1 will be described. In S506, the learning unit 213 treats the Nth category as a subcategory of the first category and the second category. Therefore, in S507, the learning unit 213 calculates the loss by treating the data of the Nth category as a correct answer regardless of whether it is identified as either the first category or the second category in S505. On the other hand, when the data of the Nth category is identified as any one of the third category to the Mth category, the learning unit 213 corrects it so that the loss increases, assuming that the error during summary application is unacceptable, and calculates the loss so as not to misidentify. For example, the learning unit 213 may multiply the calculated loss by a correction coefficient of 1 or more to increase it, or correct it so that the loss increases as the fitness decreases according to the magnitude of the fitness.

[0053] In this embodiment, the category identification unit 203 identifies the category of the input document 701 using the category identification model obtained by learning in this way. When it is identified as the first category or the first subcategory, the analysis application unit 704 controls to apply the first category summary model 707 to the input document 701. Also, when it is identified as the second category or the second subcategory, the analysis application unit 704 controls to apply the second category summary model 708 to the input document 701.

[0054] In this embodiment, when only the summary model for a specific category exists, even when a document of a category other than the specific category is obtained, by selecting the summary model for the specific category according to the fitness, it becomes possible to perform a high-quality summary task. Also, by using the document of a category other than the specific category as teacher data for learning the category identification model to identify a document of the specific category with a high fitness for that category, the category suitable for applying the summary task is easily identified. That is, by effectively utilizing the existing analysis model, it becomes possible to perform a highly accurate analysis task for input data of categories other than the specific category.

[0055] In this manner, when only an analytical model for a specific category exists, it is possible to perform highly accurate analysis of categories other than the specific category without newly learning analytical models for categories other than the specific category, thereby effectively increasing the types of categories that can be applied to the analytical model.

[0056] [Modification of the second embodiment] In this embodiment, the learning unit 213 learned the category identification model by treating the Nth category as a subcategory of the first category and a subcategory of the second category. However, the category identification model learned in this way may have a reduced identification performance for the first or second category. In such a case, the teacher dataset generation unit 215 adds a category and labels the Nth category itself, rather than labeling the subcategory in the teacher data of the Nth category. The learning unit 213 learns using the teacher dataset 212 of the Nth category thus generated. For the document 701 identified as the Nth category by the category identification model, the analysis application unit 704 applies a summary model for the category (summary model of the first or second category) that has the highest compatibility with the Nth category to summarize the document of the Nth category.

[0057] Other embodiments Although the present invention has been described above with reference to the embodiments, the above embodiments are merely illustrative of the specific examples of the present invention, and the technical scope of the present invention should not be interpreted as being limited by these embodiments. In other words, the present invention can be embodied in various forms without departing from its technical concept or main features.

[0058] The present invention can also be realized by a process in which a program for implementing one or more of the functions of the above-described embodiments is supplied to a system or device via a network or a storage medium, and one or more processors in a computer of the system or device read and execute the program. The present invention can also be realized by a circuit (e.g., ASIC) that implements one or more of the functions.

[0059] The disclosure of each of the above embodiments includes the following configurations, methods, and programs. (Configuration 1) Identification means for identifying the category of input data, Control means for selecting, from among the analysis models for each category, an analysis model to be applied to the input data identified as the first category by the identification means, based on the degree of fitness with respect to the analysis models for each category of the data belonging to the first category; An information processing apparatus characterized by comprising the same. (Configuration 2) The control means applies the analysis model for the first category to the input data identified as the first category, if the analysis model for the first category exists, and applies the selected analysis model if the analysis model for the first category does not exist. The information processing apparatus according to Configuration 1, characterized by this. (Configuration 3) The control means selects the analysis model with the highest degree of fitness from among the analysis models for each category. The information processing apparatus according to Configuration 1 or 2, characterized by this. (Configuration 4) Fitness calculation means for calculating the degree of fitness based on the comparison result between the result obtained by inputting the data belonging to the first category into the analysis models for each category and the correct information of the data belonging to the first category; The information processing apparatus according to any one of Configurations 1 to 3, further characterized by comprising the same. (Configuration 5) Learning means for learning the identification model used by the identification means, further comprising, The learning means performs learning of the identification model using teacher data in which the first category is labeled for the data belonging to the first category, When there is no analysis model for the first category, the control means applies the selected analysis model to the input data identified as the first category. The information processing apparatus according to any one of Configurations 1 to 4, characterized in that. (Configuration 6) The identification means identifies the category of the image, The analysis model is a model for detecting specific parts of a category from an image. The information processing apparatus according to any one of Configurations 1 to 5, characterized in that. (Configuration 7) The identification means identifies the category of the document, The analysis model is a model for summarizing documents of a category. The information processing apparatus according to any one of Configurations 1 to 6, characterized in that. (Configuration 8) The analysis model is a neural network. The information processing apparatus according to any one of Configurations 1 to 7, characterized in that. (Configuration 9) The identification model is a neural network. The information processing apparatus according to Configuration 5, characterized in that. (Configuration 10) It has learning means for learning an identification model for identifying the category of data, The learning means labels the data belonging to the first category with a category based on the degree of fitness of each category's analysis model for analyzing the data of each category belonging to the first category, and uses the labeled teacher data for learning the identification model. The information processing apparatus, characterized in that. (Configuration 11) Identification means for identifying the category of input data using the identification model learned by the learning means, Control means for applying the analysis model for the category identified by the identification means to the input data, The information processing apparatus according to Configuration 10, characterized by having. (Configuration 12) The learning means uses, for learning of the identification model, teacher data obtained by labeling, for data belonging to the first category, a second category which is the category of an analysis model with a fitness equal to or higher than a threshold value. The information processing apparatus according to configuration 10 or 11, characterized in that. (Configuration 13) The learning means uses, for learning of the identification model, teacher data obtained by labeling, for data belonging to the first category, a category different from the category of an analysis model with a fitness less than a threshold value. The information processing apparatus according to any one of configurations 10 to 12, characterized in that. (Configuration 14) When the identification performance of the identification model for data belonging to the second category deteriorates as a result of learning, the learning means performs learning of the identification model using teacher data obtained by labeling, for data belonging to the first category, the first category. The information processing apparatus according to configuration 12, characterized in that. (Configuration 15) When, as an identification result of the identification model for data belonging to the first category, a category of an analysis model with a fitness less than a threshold value is output, the learning means corrects so that the loss increases. The information processing apparatus according to any one of configurations 10 to 14, characterized in that. (Configuration 16) When there are a plurality of analysis models with a fitness equal to or higher than a threshold value, the learning means labels the category of the analysis model with a higher fitness. The information processing apparatus according to configuration 12, characterized in that. (Configuration 17) The identification model is a neural network. The information processing apparatus according to any one of configurations 10 to 16, characterized in that. (Configuration 18) The identification model identifies the category of an image, The analysis model is a model that detects a specific part of a category from an image. The information processing apparatus according to any one of configurations 10 to 17, characterized in that. (Configuration 19) The identification model identifies the category of a document, The information processing apparatus according to any one of configurations 10 to 18, wherein the analysis model is a model that summarizes documents in a category. (Configuration 20) The information processing apparatus according to any one of configurations 10 to 19, wherein the analysis model is a neural network. (Method) An identification step of identifying the category of input data, A control step of selecting, from among the analysis models for each category, an analysis model to be applied to the input data identified as the first category in the identification step, based on the degree of fitness of the data belonging to the first category to the analysis models for each category for analyzing the data in each category; An information processing method characterized by including the above. (Program) A program for causing a computer of an information processing apparatus to function as an identification means for identifying the category of input data, and a control means for selecting, from among the analysis models for each category, an analysis model to be applied to the input data identified as the first category by the identification means, based on the degree of fitness of the data belonging to the first category to the analysis models for each category for analyzing the data in each category. ​

Claims

1. It has a learning means for learning an identification model for identifying data categories, The learning means uses, for learning the identification model, teacher data obtained by labeling a second category, which is a category of an analysis model for which the degree of fitness of data belonging to a first category with respect to an analysis model for each category for analyzing data of each category is equal to or greater than a threshold value. An information processing apparatus characterized by this.

2. A learning means for learning an identification model for identifying data categories, the learning means using, for learning the identification model, teacher data obtained by labeling a second category, which is a category of an analysis model for which the degree of fitness of data belonging to a first category with respect to an analysis model for each category for analyzing data of each category is equal to or greater than a threshold value. Identification means for identifying the category of input data using the identification model learned by the learning means; Control means for applying an analysis model for the category identified by the identification means to the input data; An information processing apparatus characterized by having this.

3. The learning means uses, for learning the identification model, teacher data obtained by labeling data belonging to the first category with a category different from the category of an analysis model for which the degree of fitness is less than the threshold value. The information processing apparatus according to claim 2, characterized by this.

4. When the learning means determines that the identification performance of the identification model for data belonging to the second category of the identification model has deteriorated as a result of learning, the learning means uses teacher data obtained by labeling data belonging to the first category with the first category to perform learning of the identification model. The information processing apparatus according to claim 1, characterized by this.

5. When, as an identification result of the data belonging to the first category by the identification model, a category of an analysis model for which the degree of fitness is less than the threshold value is output, the learning means corrects the loss so as to increase. The information processing apparatus according to claim 1, characterized by this.

6. The learning means labels the category of the analysis model with a higher fitness when there are multiple analysis models with a fitness equal to or higher than a threshold value, in the information processing apparatus according to claim 1.

7. The identification model is a neural network, in the information processing apparatus according to claim 1.

8. The identification means identifies the category of an image, The analysis model is a model that detects a specific part of a category from an image, in the information processing apparatus according to claim 2.

9. The identification means identifies the category of a document, The analysis model is a model that summarizes a document of a category, characterized in the information processing apparatus according to claim 2.

10. A learning step in which learning means learns an identification model for identifying the category of data, and teacher data obtained by labeling a second category, which is the category of an analysis model for which the fitness of data belonging to a first category with respect to each analysis model for each category for analyzing data of each category is equal to or higher than a threshold value, is used for learning the identification model, in an information processing method of an information processing apparatus.

11. A learning means in which identification means learns an identification model for identifying the category of data, and an identification step of identifying the category of input data using the identification model learned by the learning means, where teacher data obtained by labeling a second category, which is the category of an analysis model for which the fitness of data belonging to a first category with respect to each analysis model for each category for analyzing data of each category is equal to or higher than a threshold value, is used for learning the identification model; and A control step in which control means applies an analysis model for the category identified in the identification step to the input data, characterized in an information processing method of an information processing apparatus.

12. A computer of an information processing apparatus, A learning means for learning an identification model that identifies data categories, which uses teacher data in which a second category, which is a category of an analysis model for which the degree of fitness of data belonging to a first category with respect to an analysis model for each category for analyzing data of each category is equal to or greater than a threshold value, for learning the identification model. A program for causing it to function as such.

13. A computer of an information processing apparatus, An identification means for identifying the category of input data, using the identification model learned by the learning means that is a learning means for learning an identification model that identifies data categories, which uses teacher data in which a second category, which is a category of an analysis model for which the degree of fitness of data belonging to a first category with respect to an analysis model for each category for analyzing data of each category is equal to or greater than a threshold value, for learning the identification model; A control means for applying an analysis model for the category identified by the identification means to the input data, A program for causing it to function as such.

Images