JP7507733B2 - Information processing device, information processing method, and information processing program - Google Patents

Description

The present invention relates to an information processing device, an information processing method, and an information processing program.

Various technologies have been provided for generating data used to train a model using machine learning and other methods. For example, a technology has been provided for generating training data including character string images and correct answer labels based on a database in which multiple words written in handwritten character areas of forms are registered, and a dataset of handwritten character images (for example, Patent Document 1, etc.).

Japanese Patent No. 6590355

However, there is room for improvement in the above conventional techniques. For example, the conventional techniques target handwritten character images, i.e., images, and it is difficult to generate data to be used for generating models that target character information. Therefore, it is desirable to efficiently generate character information that can be used for model training.

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 efficiently generate character information that can be used for model learning.

The information processing device according to the present application is characterized by comprising: an acquisition unit that acquires a learning dataset including first character information labeled for use in training a model that extracts a target string of extraction, which is a string corresponding to a predetermined type, from character information, and second character information that is character information without the label; and a generation unit that selects the first character information similar to the second character information from the learning dataset as similar character information, and changes the first character string that is the target string of extraction in the similar character information to the second character string that is the target string of extraction in the second character information, thereby generating changed character information that includes the second character string and is character information that can be used for training the model.

According to one aspect of the embodiment, it is possible to efficiently generate character information that can be used for model training.

FIG. 1 is a diagram illustrating an example of processing by an information processing system according to an embodiment. FIG. 2 is a diagram illustrating an example of the generation process. FIG. 3 is a diagram illustrating an example of vector conversion processing. FIG. 4 is a diagram illustrating an example of the configuration of an information processing device according to an embodiment. FIG. 5 is a diagram illustrating an example of a learning data storage unit according to the embodiment. FIG. 6 is a diagram illustrating an example of a model information storage unit according to the embodiment. FIG. 7 is a flowchart illustrating an example of processing by the information processing device according to the embodiment. FIG. 8 is a diagram illustrating an example of a hardware configuration.

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

(Embodiment)
[1. Information Processing]
An example of information processing according to the embodiment will be described with reference to Fig. 1. Fig. 1 is a diagram showing an example of processing by an information processing system according to the embodiment. In Fig. 1, a case where a model M1 that extracts named entities is generated using learning data generated by an information processing device 100 will be described as an example.

First, the configuration of the information processing system 1 will be described. As shown in FIG. 1, the information processing system 1 includes a terminal device 10 and an information processing device 100. The terminal device 10 and the information processing device 100 are connected to each other via a predetermined communication network (not shown) so as to be able to communicate with each other via a wired or wireless connection. Note that the information processing system 1 shown in FIG. 1 may include multiple terminal devices 10 and multiple information processing devices 100.

The information processing device 100 is a computer that converts a character string in labeled character information (hereinafter also referred to as "first character information") with a character string in unlabeled character information (hereinafter also referred to as "second character information") to generate character information that can be used for model learning. The information processing device 100 converts a character string that corresponds to a predetermined type in the first character information (hereinafter also referred to as "first character string") into a character string that is estimated to correspond to the predetermined type in the second character information (hereinafter also referred to as "second character string"). In the example of FIG. 1, a case where the predetermined type is a named entity will be described as an example.

As a result, the information processing device 100 generates character information (hereinafter also referred to as "changed character information") that includes the second character string and can be used for learning a model. In FIG. 1, the information processing device 100 generates changed character information to be used as new learning data, using learning data that includes first character information, such as manually labeled training data (manually labeled training data) by a specific editor ED1.

The terminal device 10 is a device (computer) that is used to assign labels (correct answer information) to data (text information) and manually generate training data (learning data). The terminal device 10 is used by a specific editor ED1 who assigns labels to data. The terminal device 10 is realized, for example, by a smartphone, a tablet terminal, a notebook PC (Personal Computer), a desktop PC, a mobile phone, a PDA (Personal Digital Assistant), or the like. FIG. 1 shows the case where the terminal device 10 is a desktop PC.

For example, a specific editor ED1 operates the terminal device 10 to assign a label to each piece of character information. For example, in response to the operation of the specific editor ED1, the terminal device 10 assigns information (label) indicating the type (attribute) to which each character string in the character information corresponds as correct answer information, such as the first character information LD1, which is the labeled data shown in FIG. 2.

For example, the terminal device 10 adds a label to the character information indicating the position (range) where a character string corresponding to a predetermined type is included in the character information. For example, a label indicating a character string corresponding to a proper noun (named entity) such as an organization name is added to the character information. The terminal device 10 generates learning data by adding a label to the character information in response to the operation of a predetermined editor ED1. The terminal device 10 transmits the manually generated training data (learning data) to the information processing device 100.

The information processing system 1 may also include a user terminal, which is a device used by a user. The user terminal may be, for example, a smartphone, a tablet terminal, a notebook PC, a desktop PC, a mobile phone, a PDA, or the like, and provides various services to the user. The user terminal executes processing and displays information in response to user operations.

The information processing system 1 may also include a service providing device that provides a user with a service related to an Internet encyclopedia. Note that the Internet encyclopedia referred to here may be, for example, Wikipedia. The service providing device provides information related to commentary content that explains a specific subject. In response to a request from the information processing device 100, the service providing device transmits information about the commentary content that explains the specific subject to the information processing device 100. In response to a request from the user terminal, the service providing device transmits information about the commentary content that explains the specific subject to the user terminal.

An example of information processing will be described below with reference to FIG. 1. First, a predetermined editor ED1 assigns a label to each piece of character information (step S11). For example, the predetermined editor ED1 checks the content of the unlabeled character information, and generates manually labeled training data by assigning a label indicating that each character string in the character information corresponds to a named entity to the character string, which indicates that the character string is a named entity. For example, the predetermined editor ED1 may operate the terminal device 10 to assign information (labels) indicating the type (attribute) to which each character string in the character information corresponds as correct answer information, such as the first character information LD1, which is the labeled data shown in FIG. 2; details of FIG. 2 will be described later.

After the specified editor ED1 has finished labeling each piece of character information, the specified editor ED1 operates the terminal device 10 to transmit the manually labeled training data to the information processing device 100 (step S12). In response to the operation of the specified editor ED1, the terminal device 10 transmits the manually labeled training data to which the specified editor ED1 has added labels to the information processing device 100.

As a result, the information processing device 100 acquires the manually-labeled training data. Then, the information processing device 100 adds the acquired manually-labeled training data to the learning dataset DS1 as first character information. Specifically, the information processing device 100 registers the manually-labeled training data received from the terminal device 10 in the learning data storage unit 121 (see FIG. 5) as data to be used for learning.

Then, the information processing device 100 performs a process of generating new character information using the first character information group in the learning dataset DS1. In FIG. 1, the information processing device 100 performs a process of generating new character information using the second character information UD1 extracted from the content as a target. Here, the second character information UD1 is character information to which no label is attached. For example, the second character information UD1 is character information included in a content that describes a predetermined target (in the example of FIG. 2, "XX on the Xth of the week"), like the second character information UD1 shown in FIG. 2. Note that the content to be extracted from the second character information UD1 may be various contents such as contents provided in an Internet encyclopedia. The second character information UD1 includes "XX on the Xth of the week" as the second character string of the named expression. Note that although XX on the Xth of the week is abstractly shown, XX on the Xth of the week is a proper noun (proper name) that exists and is a neologism that indicates a newly appeared artist (organization name).

The information processing device 100 selects first character information similar to the second character information UD1 from the learning dataset DS1 (step S13). For example, the information processing device 100 may vectorize each piece of character information and select similar character information based on the similarity of the vectors. In this case, the information processing device 100 selects similar character information based on the similarity between each of the first vectors obtained by vectorizing each piece of first character information in the learning dataset DS1 and the second vector obtained by vectorizing the second character information UD1.

For example, the information processing device 100 selects, as similar character information, the first character information corresponding to the first vector having the maximum cosine similarity with the second vector. Note that the similarity relationship between the vectors is not limited to cosine similarity, and any index may be used, for example, Euclidean distance or Mahalanobis distance. For example, the information processing device 100 converts each piece of character information into a vector using a model M2 that converts character information into a vector. For example, the information processing device 100 converts each piece of character information into a vector using a model M2 that has been trained using various techniques related to word2vec (also called "w2v").

For example, the information processing device 100 inputs each character string corresponding to a noun in the second character information UD1 into the model M2, and sets the average of the vectors output by the model M2 as the vector of the second character information UD1 (second vector). The information processing device 100 also inputs each character string corresponding to a noun in the first character information LD1 into the model M2, and sets the average of the vectors output by the model M2 as the vector of the first character information LD1 (first vector). Note that the above is merely an example, and the information processing device 100 may set the second vector as a vector obtained by vector-converting the entire second character information UD1, and the first vector as a vector obtained by vector-converting the entire first character information LD1.

In FIG. 1, the information processing device 100 selects, from the learning dataset DS1, the first character information LD1 that has the highest similarity to the second character information UD1 as similar character information. The first character information LD1 includes "Artist A" as the first character string of the named entity.

Although artist A is referred to abstractly, artist A is assumed to be an existing proper noun (proper name). Furthermore, if there is no first character information similar to second character information UD1, information processing device 100 may exclude second character information UD1 from processing. For example, if there is no first character information whose similarity to second character information UD1 is equal to or greater than a predetermined value, information processing device 100 may exclude second character information UD1 from processing.

Then, the information processing device 100 converts the first character string in the similar character information into the second character string, thereby generating changed character information that includes the second character string and can be used for learning the model (step S14). In FIG. 1, the information processing device 100 converts the named entity "artist A" in the first character information LD1 into the named entity "XX on the Xth day of the week" in the second character information UD1. As a result, the information processing device 100 generates changed character information CD1 in which the first character string "artist A" in the first character information LD1 is converted into the second character string "XX on the Xth day of the week". In other words, the information processing device 100 generates changed character information CD1 that includes the second character string "XX on the Xth day of the week" and can be used for learning the model M1.

The information processing device 100 adds the generated changed character information to the data used for learning as first character information (step S15). In FIG. 1, the information processing device 100 adds changed character information CD1, in which the first character string "Artist A" in the first character information LD1 is converted to the second character string "XX on the Xth day of the week", to the learning data set DS1. For example, the information processing device 100 associates the changed character information CD1 including "XX on the Xth day of the week" with a label indicating that "XX on the Xth day of the week" in the changed character information CD1 is a named entity, and stores the result in the learning data storage unit 121 as the first character information.

Then, the information processing device 100 uses the training data set DS1 to which the changed character information CD1 has been added to train the model M1 (step S16). The information processing device 100 uses the training data set DS1 to train (update) parameters such as weights of the model M1. Any method can be used for the training process of the model M1.

For example, when the character information input to the model M1 includes a named entity, the information processing device 100 trains the model M1 to output information indicating the named entity. For example, when the changed character information CD1 is input to the model M1, the information processing device 100 trains the model M1 to output the character string "XX on the Xth day of the week". For example, the model M1 may be a network such as a recurrent neural network (RNN) or LSTM (Long Short-Term Memory units), which is an extension of the RNN. Note that the above is merely an example, and the model M1 may adopt any network configuration other than a recurrent neural network as long as it is capable of extracting a predetermined type of character string from character information.

As described above, the information processing device 100 can efficiently generate character information that can be used for model training by converting the first character string of already labeled data (first character information) into another character string (second character string) to generate new training data. In the example of FIG. 1, the information processing device 100 can add character information including the new word "XX on the Xth day of the week" to the training data, so that a model that is likely to properly extract a named entity even if the named entity is a new word can be trained. In this way, if the training data contains sufficient examples of such new words, the machine learning model is more likely to be able to deal with these new words, so in the information processing system 1, training data including the named entity of the new word is generated and padded, so that a model that is likely to properly extract even a new word can be trained.

[1-1. Processing example]
Here, a processing example related to the content described in Fig. 1 will be described with reference to Fig. 2 and Fig. 3. Fig. 2 is a diagram showing an example of a generation process. Fig. 3 is a diagram showing an example of a vector conversion process.

As shown in FIG. 2, the second character information UD1 is character information extracted from the content CT of an Internet encyclopedia such as Wikipedia. For newly created named entities such as neologisms, the information processing system 1 often does not have sufficient information on hand. For this reason, the information processing device 100 collects, for example, entries with a large number of links that were recently added to the Internet encyclopedia (for example, within the last month). In this way, the information processing device 100 obtains, for example, explanatory content from the Internet encyclopedia.

Then, the information processing device 100 estimates named entities from the character strings contained in the commentary content from the tags attached to the commentary content. For example, the information processing device 100 may analyze the HTML (Hyper Text Markup Language) of the commentary content and estimate various pieces of information. For example, the information processing device 100 analyzes the HTML of the commentary content and estimates that the heading part of the commentary content is the subject that the commentary content explains.

In addition, for example, when the tag attached to the commentary content is the name of an organization such as a music unit, the information processing device 100 may infer that the object indicated by the heading portion of the commentary content is a named expression. For example, when the category indicated by the tag attached to the commentary content is the name of an organization such as a music unit, the information processing device 100 may infer that the object indicated by the heading portion of the commentary content is a named expression.

The information processing device 100 may then extract the character string indicated by the heading of the commentary content as the second character string together with the sentence following the heading in the commentary content as the second character information. Note that the above is merely an example, and the second character information may be information including various information. Furthermore, the information processing device 100 may obtain the second character information extracted from the analysis content from the service providing device.

For example, the content CT of an Internet encyclopedia is large, for example 50,000 entries, and the cost of collecting it is low. On the other hand, the cost of generating a learning data set DS1 that includes manually labeled training data is high, and the number of entries is small, for example 3,000. Therefore, the information processing device 100 automatically generates new learning data using the content CT and the manually labeled training data MD1.

For example, the information processing device 100 selects, from among the commentary content included in the content CT, content in which the subject to be explained is a new word (also referred to as "new word content"). For example, the information processing device 100 estimates that, from among the subjects explained by each commentary content, an object that is not included as a named entity in the first character information in the learning dataset DS1 is a new word. Then, the information processing device 100 selects, as new word content, the content in which the subject to be explained is estimated to be a new word. The information processing device 100 generates second character information from the new word content.

For example, the information processing device 100 generates the second character information by extracting a character string (second character string) indicated by the heading portion of the commentary content and the sentence following the heading portion in the commentary content. The information processing device 100 generates the second character information UD1 by extracting the second character string "XX on the Xth day of the week" indicated by the heading portion from commentary content with the heading "XX on the Xth day of the week" and the sentence following the heading portion in the commentary content.

Then, the information processing device 100 selects, from the learning dataset DS1, first character information that is similar to the second character information UD1. For example, the information processing device 100 vectorizes each piece of first character information and second character information UD1 in the learning dataset DS1, and selects similar character information based on the similarity of the vectors. The information processing device 100 vectorizes each piece of first character information and second character information UD1 in the learning dataset DS1. The information processing device 100 converts each piece of character information into a vector using a model M2 that converts character information into a vector.

In FIG. 3, the information processing device 100 converts the second character information UD1 into a vector by inputting the second character information UD1 into a model M2 and having the model M2 output a vector VC1 obtained by vectorizing the second character information UD1. The information processing device 100 also converts the first character information LD1 into a vector by inputting the first character information LD1 into the model M2 and having the model M2 output a vector VC2 obtained by vectorizing the first character information LD1. Note that while FIG. 3 illustrates only the first character information LD1, the information processing device 100 vectorizes each piece of first character information in the training dataset DS1.

For example, the information processing device 100 selects, as similar character information, the first character information corresponding to the vector having the maximum cosine similarity with the vector VC1 of the second character information UD1. In FIG. 2, the information processing device 100 selects, as similar character information, the first character information LD1 containing "artist A" as the first character string of the named entity. In this way, the information processing device 100 executes processing on entries having a description (vector) most similar to a sentence (vector) in the learning data. This allows the information processing device 100 to reduce the possibility of selecting an entry unrelated to the original sentence. This allows the information processing device 100 to execute processing on entries having a description similar to a sentence in the learning data.

Then, the information processing device 100 converts the named entity "artist A" in the first character information LD1 into the named entity "XX on the Xth day of the week" in the second character information UD1. As a result, the information processing device 100 generates changed character information CD1 in which "artist A", the first character string in the first character information LD1, is converted into the second character string "XX on the Xth day of the week". In the example of FIG. 2, the information processing device 100 generates changed character information CD1 in which "XX on the Xth day of the week" is placed in the label portion of the organization name, which is an example of a named entity. As a result, the information processing device 100 can train a model using new learning data that has been automatically (artificially) created.

Note that the above process is merely an example, and the information processing device 100 may perform any process as long as it is capable of generating changed character information. For example, the information processing device 100 may select one piece of first character information from the learning dataset DS1, and select second character information similar to the selected first character information (selected first character information) from the content CT. In this case, if there is no second character information similar to the selected first character information, the information processing device 100 may exclude the selected first character information from the processing target. For example, if there is no second character information whose similarity to the selected first character information is equal to or greater than a predetermined value, the information processing device 100 may exclude the selected first character information from the processing target.

[1-2. Other examples]
The above-described process is merely an example, and the information processing system 1 may perform various processes. For example, in the above-described process, a case where a model that extracts named entities is learned is shown, but the model to be learned is not limited to a model that extracts named entities. For example, the model may be a model that extracts keywords from text such as an input article, finds themes and names of artifacts, and provides links to wikis, etc. Also, for example, the model may be a model used to mask personal information such as people's names and credit card numbers. Any model can be adopted as long as it is a model for performing keyword-related processing.

For example, the information processing system 1 may expand the learning data by finding a string similar to the string to be replaced and replacing it. For example, the information processing system 1 may target content with high link similarity in an Internet encyclopedia or content with similar article content in an Internet encyclopedia. For example, the information processing system 1 may vectorize only nouns using any vectorization technology such as w2v or s2v, and average them. The information processing system 1 may also create a summary and vectorize it.

For example, the information processing system 1 may generate new changed character information by extracting and rearranging each word (organization name, artifact, etc.) in the training data from the same page of the Internet encyclopedia. Furthermore, the character information may contain explanatory text, and the original training data may be different from the data it replaces. Furthermore, the information processing system 1 may create a model for each category to which it is desired to apply. For example, the information processing system 1 may generate learning data for each category of the Internet encyclopedia, such as Japanese music groups, and generate a model for each category using the learning data for each category.

2. 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.

(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 predetermined communication network by wire or wirelessly, and transmits and receives information to and from the terminal device 10.

(Memory 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, a model information storage unit 122, and a content information storage unit 123.

(Learning Data Storage Unit 121)
The learning data storage unit 121 according to the embodiment stores various information related to data used for learning. The learning data storage unit 121 stores learning data (dataset) used for learning. FIG. 5 is a diagram illustrating an example of the learning data storage unit according to the embodiment. For example, the learning data storage unit 121 stores various information related to various data such as learning data used for learning and evaluation data used for accuracy evaluation (measurement). FIG. 5 illustrates an example of the learning data storage unit 121 according to the embodiment. In the example of FIG. 5, the learning data storage unit 121 includes items such as "dataset ID", "data ID", "data", "label", and "date and time".

"Dataset ID" indicates identification information for identifying a dataset. "Data ID" indicates identification information for identifying data. Furthermore, "Data" indicates data corresponding to the data identified by the data ID.

A "label" refers to a label (correct label) that is attached to the corresponding data. For example, a "label" may be information (correct answer information) that indicates which type each character string in the corresponding data (character information) corresponds to. For example, a "label" is correct answer information that indicates the position (range) in the character information where a character string that corresponds to a specific type is included. For example, a "label" may be correct answer information that indicates a character string in the character information that corresponds to a proper noun (named entity) such as an organization name. In addition, a "label" may be correct answer information that indicates a character string that corresponds to a named entity such as a proper noun such as a person's name or place name, a date expression, or a time expression.

In addition, "date and time" indicates the time (date and time) related to the corresponding data. Note that in the example of FIG. 5, "DA1" and the like are illustrated, but "date and time" may be a specific date and time such as "August 12, 2021, 17:48:37", or information indicating which model learning the data started to be used from, such as "Use started from model learning of version XX", may be stored.

The example in FIG. 5 shows that the dataset (dataset DS1) identified by the dataset ID "DS1" includes multiple data identified by data IDs "DID1", "DID2", "DID3", etc. For example, each data (learning data) identified by the data IDs "DID1", "DID2", "DID3", etc. is character information (character data) used for model training, etc.

For example, data DT1 identified by data ID "DID1" is labeled data with label LB1, indicating that its use began with model training at date and time DA1. Furthermore, for example, data DT4 identified by data ID "DID4" is unlabeled data collected and with label LB4, which is a predictive label, indicating that its use began with model training at date and time DA4.

The learning data storage unit 121 may store various information according to the purpose, without being limited to the above. For example, the learning data storage unit 121 may store data in a manner that allows each piece of data to be identified as learning data or evaluation data. For example, the learning data storage unit 121 stores learning data and evaluation data in a manner that allows them to be distinguished. The learning data storage unit 121 may store information that identifies whether each piece of data is learning data or evaluation data. The information processing device 100 learns a model based on each piece of data used as learning data and the correct answer information. The information processing device 100 calculates the accuracy of the model based on each piece of data used as evaluation data and the correct answer information. The information processing device 100 calculates the accuracy of the model by collecting a result of comparing the output result output by the model when the evaluation data is input with the correct answer information.

(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) of a trained model (model) trained (generated) by a learning process. Fig. 6 is a diagram illustrating an example of a model information storage unit according to the embodiment. In the example shown in Fig. 6, the model information storage unit 122 includes items such as "model ID", "purpose", and "model data".

"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 6 etc. shows an example in which conceptual information such as "MDT1" is stored in "model data", but in reality, various information that constitutes 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.

In the example shown in FIG. 6, the model identified by the model ID "M1" (model M1) indicates that its use is "named entity extraction." In other words, model M1 indicates that it is a model that outputs information (such as a character string) that indicates a character string that corresponds to a named entity in input character information. In addition, it indicates that the model data of model M1 is model data MDT1.

The model (model M2) identified by the model ID "M2" indicates that its use is "vector conversion." In other words, model M2 indicates that it is a model that outputs vectors obtained by vector-converting input information (e.g., character information). The model data of model M2 indicates that it is model data MDT2.

The model information storage unit 122 may store various types of information depending on the purpose, not limited to the above.

(Content information storage unit 123)
The content information storage unit 123 according to the embodiment stores various information related to the content. For example, the content information storage unit 123 stores various information related to the content from which the second character information is extracted. For example, the content information storage unit 123 stores information on a specific content provided on the Internet. For example, the content information storage unit 123 stores information on an explanation content that explains a specific subject. For example, the content information storage unit 123 stores information on a content in an Internet encyclopedia. For example, the content information storage unit 123 stores information on a content received from a service providing device that provides a user with a service related to the Internet encyclopedia.

The content information storage unit 123 stores second character information extracted from specified content. The content information storage unit 123 stores second character information extracted from specified content provided on the Internet. The content information storage unit 123 stores second character information extracted from commentary content that explains a specified subject. The content information storage unit 123 stores second character information extracted from commentary content that explains a subject different from the subject indicated by the first character string. The content information storage unit 123 stores second character information extracted from content in an Internet encyclopedia.

Note that the above is merely an example, and the content information storage unit 123 may store information on various contents, etc.

(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), an MPU (Micro Processing Unit), or the like, executing various programs (corresponding to an example of an information processing program) 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).

As shown in FIG. 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 the 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. Also, the connection relationship between each processing unit of the control unit 130 is not limited to the connection relationship shown in FIG. 4, and may be other connection relationships.

(Acquisition unit 131)
The acquisition unit 131 receives various information from an external information processing device via the communication unit 110. The acquisition unit 131 receives various information from the terminal device 10. The acquisition unit 131 stores the labeled training data (character information) received from the terminal device 10 in the storage unit 120 as learning data. The acquisition unit 131 registers the labeled character information received from the terminal device 10 in the learning data storage unit 121 as data used for learning a model (learning data). The acquisition unit 131 also receives content information from a service providing device that provides a service related to an Internet encyclopedia to a user. The acquisition unit 131 may receive second character information from the service providing device.

The acquisition unit 131 acquires various types of information from the storage unit 120. The acquisition unit 131 acquires data to be used for learning from the learning data storage unit 121. The acquisition unit 131 acquires model information from the model information storage unit 122.

The acquisition unit 131 acquires second character information extracted from specified content. The acquisition unit 131 acquires second character information extracted from specified content provided on the Internet. The acquisition unit 131 acquires second character information extracted from commentary content that explains a specified object. The acquisition unit 131 acquires second character information extracted from commentary content that explains an object different from the object indicated by the first character string. The acquisition unit 131 acquires second character information extracted from content in an Internet encyclopedia.

(Generation unit 132)
The generation unit 132 generates various information. For example, the generation unit 132 generates various information (data) from information (data) stored in the storage unit 120. The generation unit 132 registers the generated information in the storage unit 120. For example, the generation unit 132 generates various information from information (data) stored in the learning data storage unit 121, the model information storage unit 122, the content information storage unit 123, etc.

The generation unit 132 selects various information. The generation unit 132 selects character information that satisfies a predetermined condition from the training dataset. The generation unit 132 selects similar character information from the training dataset based on the similarity with the second character information. The generation unit 132 selects, from the training dataset, first character information that has the greatest similarity with the second character information as similar character information. The generation unit 132 selects similar character information based on the similarity between each of first vectors obtained by vectorizing each piece of first character information in the training dataset and a second vector obtained by vectorizing the second character information.

The generation unit 132 estimates various information. The generation unit 132 estimates new words from character strings included in the character information. For example, the generation unit 132 estimates that, among the objects explained by each explanatory content, an object that is not included as a named entity in the first character information in the learning dataset DS1 is a new word. The generation unit 132 then selects the content in which the object to be explained is estimated to be a new word as new word content. The generation unit 132 generates second character information from the new word content.

The generating unit 132 generates changed character information by changing a first character string in the similar character information to a second character string in the second character information. The generating unit 132 generates changed character information in which a type label indicating that the first character string corresponds to a predetermined type is set as a type label of the second character string. The generating unit 132 generates changed character information by changing a first character string that is a named entity in the similar character information to a second character string that is a named entity in the second character information.

(Learning Unit 133)
The learning unit 133 learns a model. The learning unit 133 learns various pieces of information based on information from an external information processing device and information stored in the storage unit 120. The learning unit 133 learns various pieces of information based on information stored in the learning data storage unit 121. The learning unit 133 stores the model generated by learning in the model information storage unit 122.

The learning unit 133 learns a model using learning data including character information generated by the generation unit 132. The learning unit 133 learns a model by a machine learning process using the changed character information generated by the generation unit 132. In response to input of character information, the learning unit 133 learns a model that outputs a character string to be extracted when the character information includes the character string to be extracted. The learning unit 133 learns a model so that a second character string is extracted from the changed character information.

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 through learning. The learning unit 133 learns the model using various machine learning techniques. For example, the learning unit 133 learns the parameters of the model (network). The learning unit 133 learns the model using various machine learning techniques.

The learning unit 133 performs a learning process based on the learning data (teacher data) stored in the learning data storage unit 121. The learning unit 133 learns the parameters of the model (network). The learning unit 133 learns parameters such as the connection coefficients (weights) between connected nodes. The learning unit 133 learns the model using various machine learning techniques. The learning unit 133 learns the model parameters by a learning process performed using data to be input to the model and correct answer data indicating the output when that data is input, that is, a supervised learning method. Note that the above is just one example, and the learning unit 133 may learn the model parameters by any learning process as long as it is possible to learn the model parameters.

The learning unit 133 generates a model M1. The learning unit 133 learns the parameters of the network. For example, the learning unit 133 learns the parameters of the network of the model M1. The learning unit 133 performs a learning process using the learning data stored in the learning data storage unit 121 to generate the model M1. For example, the learning unit 133 generates a model used for named entity extraction. The learning unit 133 generates the model M1 by learning the parameters of the network of the model M1.

The method of learning by the learning unit 133 is not particularly limited, but for example, learning data that links labels with data (character information) may be prepared, and the learning data may be input to a computational model based on a multi-layer neural network for learning. The learning unit 133 may use a method based on a recurrent neural network (RNN) or an LSTM that is an extension of an RNN.

For example, the learning unit 133 may learn model M1, which is a Seq2Seq (Sequence to Sequence Model). For example, Seq2Seq is an Encoder-Decoder model whose component is an LSTM, a type of RNN. For example, when character information corresponding to the second character information UD1 in FIG. 2 is input, model M1 outputs the character string "XX on the Xth day of the week." In this way, model M1, which is Seq2Seq, performs vectorization on the Encoder side where character information corresponding to the second character information UD1 is input, and learns the RNN to output "XX on the Xth day of the week" on the Decoder side.

(Processing Unit 134)
The processing unit 134 executes various processes. The processing unit 134 executes processes using the model M1 learned by the learning unit 133. The processing unit 134 inputs character information to the model M1, and treats the character string output by the model M1 as a character string of a named entity. For example, the processing unit 134 inputs character information to the model M1, and treats the character string output by the model M1 as a new word.

The processing unit 134 inputs the character information into the model M1, and causes the providing unit 135 to transmit the character string output by the model M1, which is information indicating the character string of the named entity, to an external device.

(Providing Unit 135)
The providing unit 135 transmits information to an external device via the communication unit 110. The providing unit 135 provides an information provision service to the terminal device 10 used by the user. For example, the providing unit 135 transmits the model M1 learned by the learning unit 133 to the terminal device 10. The providing unit 135 transmits information indicating a processing result by the processing unit 134 to the terminal device 10.

The providing unit 135 provides information indicating the processing result by the processing unit 134. The providing unit 135 transmits information indicating the new word to the terminal device 10. The providing unit 135 transmits information indicating the named entity to the terminal device 10.

3. Processing flow
Next, a procedure of information processing by the information processing system 1 according to the embodiment will be described with reference to Fig. 7. Fig. 7 is a flowchart showing an example of processing by the information processing device according to the embodiment.

As shown in FIG. 7, the information processing device 100 acquires learning dataset character information including first character information labeled with a label for use in training a model that extracts a target character string, which is a character string corresponding to a predetermined type, from character information (step S101).

The information processing device 100 acquires second character information, which is unlabeled character information (step S102). The information processing device 100 selects first character information that is similar to the second character information from the learning dataset as similar character information (step S103).

The information processing device 100 changes the first character string, which is a character string to be extracted in the similar character information, to the second character string, which is a character string to be extracted in the second character information, to generate changed character information that includes the second character string and can be used for learning the model (step S104).

4. Effects
As described above, the information processing device 100 according to the embodiment includes the acquisition unit 131 and the generation unit 132. The acquisition unit 131 acquires a learning dataset including first character information labeled for use in learning a model that extracts an extraction target character string, which is a character string corresponding to a predetermined type, from character information, and second character information that is unlabeled character information. The generation unit 132 selects, from the learning dataset, first character information similar to the second character information as similar character information, and changes the first character string, which is an extraction target character string in the similar character information, to the second character string, which is an extraction target character string in the second character information, thereby generating changed character information that includes the second character string and is character information that can be used for learning the model.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by generating new training data by converting character strings of character information contained in existing training data.

In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires second character information extracted from the specified content.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by converting the first character string of the first character information into the second character string of the second character information extracted from a specific content to generate new training data.

In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires second character information extracted from specific content provided on the Internet.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by converting the first character string of the first character information into the second character string of the second character information extracted from specific content provided on the Internet to generate new training data.

In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires second character information extracted from commentary content that explains a specific subject.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by generating new learning data by converting the first character string of the first character information into the second character string of the second character information extracted from commentary content that explains a specific target.

In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires second character information extracted from commentary content that explains an object different from the object indicated by the first character string.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by converting the first character string of the first character information into the second character string of the second character information extracted from commentary content that explains an object different from the object indicated by the first character string to generate new training data.

In addition, in the information processing device 100 according to the embodiment, the acquisition unit 131 acquires second character information extracted from content in the Internet encyclopedia.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by converting the first character string of the first character information into the second character string of the second character information extracted from the content in the Internet encyclopedia to generate new training data.

In addition, in the information processing device 100 according to the embodiment, the generation unit 132 selects similar character information from the learning dataset based on the similarity with the second character information, and generates changed character information by changing the first character string in the similar character information to the second character string in the second character information.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by selecting similar character information from the training dataset based on the similarity with the second character information and generating changed character information using the selected similar character information.

In addition, in the information processing device 100 according to the embodiment, the generation unit 132 selects, from the learning dataset, the first character information that has the highest similarity to the second character information as the similar character information.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by generating changed character information using the first character information in the training dataset that has the highest similarity to the second character information.

In addition, in the information processing device 100 according to the embodiment, the generation unit 132 selects similar character information based on the similarity between each of the first vectors obtained by vectorizing each of the first character information in the learning dataset and the second vector obtained by vectorizing the second character information.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training by generating changed character information using first character information that is similar to second character information in a vectorized state from the training dataset.

In addition, in the information processing device 100 according to the embodiment, the generation unit 132 generates changed character information in which a type label indicating that the first character string corresponds to a predetermined type is set as a type label of the second character string.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used for model training, since the label of the first string is automatically assigned to the second string by using the label of the first string as the label of the second string from the training dataset.

In the information processing device 100 according to the embodiment, the acquisition unit 131 acquires a learning dataset used to train a model that extracts extraction target character strings corresponding to named entities from character information. The generation unit 132 generates changed character information by changing a first character string that is a named entity in the similar character information to a second character string that is a named entity in the second character information.

In this way, the information processing device 100 according to the embodiment can efficiently generate character information that can be used to train a model that extracts named entities by converting named entities in character information into other named entities and generating new learning data.

The information processing device 100 according to the embodiment also includes a learning unit 133. The learning unit 133 learns a model through machine learning processing using the changed character information generated by the generation unit 132.

As a result, the information processing device 100 according to the embodiment can appropriately learn a model using the generated information by learning the model through machine learning processing using the changed character information.

In addition, in the information processing device 100 according to the embodiment, the learning unit 133 learns a model that outputs an extraction target string when the character information includes an extraction target string in response to input of the character information.

As a result, the information processing device 100 according to the embodiment can learn a model that uses the generated information to output the extraction target string.

In addition, in the information processing device 100 according to the embodiment, the learning unit 133 learns a model so that the second character string is extracted from the changed character information.

As a result, the information processing device 100 according to the embodiment can learn a model that uses the generated information to output the extraction target string.

5. Hardware Configuration
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.

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, for example, by 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.

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), a HDD (Hard Disk Drive), a 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 SD (Secure Digital) 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.

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, projector, 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, keyboard, keypad, button, scanner, etc., and is realized by a USB, etc.

In addition, 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.

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

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.

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

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.

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 other devices via the network I/F 1080 and use the functions and data of a program by calling them from other programs of the other devices.

[6. Other]
Although the embodiments of the present application have been described above, the present invention is not limited to the contents of these embodiments. The above-described components include those that a person skilled in the art can easily imagine, those that are substantially the same, and those that are within the so-called equivalent range. Furthermore, the above-described components can be appropriately combined. Furthermore, various omissions, substitutions, or modifications of the components can be made without departing from the spirit of the above-described embodiments.

Furthermore, among the processes described in the above embodiments, 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 using known methods. In addition, the information including the processing procedures, specific names, various data, and parameters shown in the above documents and drawings can be changed as desired unless otherwise specified. For example, the various information shown in each drawing is not limited to the information shown in the drawings.

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 depending on various loads, usage conditions, etc.

For example, the information processing device 100 described above may be realized by multiple server computers, and depending on the functions, the configuration can be flexibly changed, such as by calling an external platform using an API (Application Programming Interface) or network computing.

The above-described embodiments and variations can be combined as appropriate to the extent that they do not cause inconsistencies in the processing content.

The above-mentioned "section, module, unit" can be read as "means" or "circuit." For example, an acquisition unit can be read as an acquisition means or an acquisition circuit.

REFERENCE SIGNS LIST 1 Information processing system 100 Information processing device 120 Storage unit 121 Learning data storage unit 122 Model information storage unit 123 Content information storage unit 130 Control unit 131 Acquisition unit 132 Generation unit 133 Learning unit 134 Processing unit 135 Provision unit 10 Terminal device

Claims (16)

an acquisition unit that acquires a learning dataset including first character information labeled with a label for use in learning a model that extracts an extraction target string, which is a string corresponding to a predetermined type, from character information, and second character information that is character information without the label and, when the predetermined type is a named entity, is character information included in content that describes an explanation of a predetermined target corresponding to the named entity ;
a generation unit that selects, from the learning dataset, the first character information similar to the second character information as similar character information , identifies the named entity of the predetermined target that is the extraction target character string in the second character information as the second character string, and generates changed character information that includes the second character string and is character information that can be used for learning the model by changing the first character string that is the extraction target character string in the similar character information to the second character string;
An information processing device comprising: The acquisition unit is
The information processing apparatus according to claim 1 , further comprising: acquiring the second character information extracted from a predetermined content. The acquisition unit is
The information processing apparatus according to claim 2 , further comprising: acquiring the second character information extracted from the predetermined content provided on the Internet. The acquisition unit is
The information processing apparatus according to claim 2 or 3, characterized in that the second character information is acquired by extracting the second character information from an explanation content that explains a predetermined subject. The acquisition unit is
The information processing apparatus according to claim 4 , further comprising: acquiring the second character information extracted from the commentary content that explains an object different from the object indicated by the first character string. The acquisition unit is
6. The information processing apparatus according to claim 2, further comprising: acquiring the second character information extracted from a content in an Internet encyclopedia. The generation unit is
The information processing device according to any one of claims 1 to 6, characterized in that the similar character information is selected from the learning dataset based on a similarity to the second character information, and the changed character information is generated by changing the first character string in the similar character information to the second character string in the second character information. The generation unit is
The information processing apparatus according to claim 7 , further comprising: selecting, from the learning data set, the first character information having a maximum similarity to the second character information as the similar character information. The generation unit is
9. The information processing apparatus according to claim 7, further comprising: selecting the similar character information based on a similarity between each of first vectors obtained by vectorizing each of the first character information in the learning dataset and a second vector obtained by vectorizing the second character information. The generation unit is
The information processing device according to any one of claims 1 to 9, characterized in that the changed character information is generated with a type label indicating that the first character string corresponds to the predetermined type as a type label of the second character string. The acquisition unit is
acquiring the learning dataset used for learning the model that extracts the extraction target character string corresponding to a named entity from character information;
The generation unit is
The information processing device according to any one of claims 1 to 10, characterized in that the changed character information is generated by changing the first character string, which is a unique expression in the similar character information, to the second character string, which is a unique expression in the second character information. a learning unit that learns the model by a machine learning process using the changed character information generated by the generation unit;
The information processing device according to any one of claims 1 to 11, further comprising: The learning unit is
The information processing apparatus according to claim 12 , further comprising: learning the model to output the extraction target character string when the extraction target character string is included in input character information. The learning unit is
The information processing apparatus according to claim 13 , further comprising: learning the model so that the second character string is extracted from the changed character information. 1. A computer-implemented information processing method, comprising:
an acquisition step of acquiring a learning dataset including first character information labeled with a label for use in learning a model that extracts an extraction target character string, which is a character string corresponding to a predetermined type, from character information, and second character information that is character information without the label and, when the predetermined type is a named entity, is character information included in content that describes an explanation of a predetermined object corresponding to the named entity;
a generating step of selecting, from the learning dataset, the first character information similar to the second character information as similar character information, identifying the named entity of the predetermined target which is the extraction target character string in the second character information as the second character string, and changing the first character string which is the extraction target character string in the similar character information to the second character string, thereby generating changed character information which includes the second character string and is character information usable for learning the model;
13. An information processing method comprising: an acquisition step of acquiring a learning dataset including first character information labeled with a label for use in learning a model that extracts an extraction target character string, which is a character string corresponding to a predetermined type, from character information, and second character information that is character information without the label and, when the predetermined type is a named entity, is character information included in content that describes an explanation of a predetermined object corresponding to the named entity;
a generation step of selecting, from the learning dataset, the first character information similar to the second character information as similar character information, identifying the named entity of the predetermined target that is the extraction target character string in the second character information as the second character string, and changing the first character string that is the extraction target character string in the similar character information to the second character string, thereby generating changed character information that includes the second character string and is character information that can be used for learning the model;
An information processing program characterized by causing a computer to execute the above.

Images