JP7410066B2 - Information provision device, information provision method, and information provision program - Google Patents

Description

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

In recent years, so-called Q&A (Q&A) has become popular, using the Internet to share knowledge and wisdom among users by having other users (answers) post answers to questions posted by users (askers). Question Answering) site is known. This type of Q&A site accumulates a large number of questions and answers (hereinafter referred to as questions and answers), so these questions and answers are used to create FAQs (Frequently Asked Questions), which are collections of questions and answers in various categories. : Frequently Asked Questions) service is desired. 2. Description of the Related Art Conventionally, a technique for creating FAQs from clustering results of a plurality of question answers is known.

JP2020-166426A

However, with the above-mentioned conventional technology, it may not be possible to say that a FAQ (question and answer collection) that meets the user's intention is provided.

For example, if there are variations in the granularity or axis (direction) of question and answers in each category, even if FAQs are created from the clustering results of these questions and answers, there will still be variations, and FAQs that match the user's intentions will not be provided. There is a possibility that it will not be done.

This application was made in view of the above, and the purpose is to provide a question and answer collection that meets the user's intentions.

The information providing device according to the present application includes a learning processing unit that performs distance learning based on user instructions for a plurality of question answers mapped to a predetermined semantic space, and a clustering process that clusters the distance learned question answers. and an FAQ creation section that summarizes each cluster including clustered questions and answers to create a question and answer collection.

According to one aspect of the embodiment, it is possible to create a question and answer collection that matches the user's intention.

FIG. 1 is a diagram showing an example of an information providing apparatus according to this embodiment. FIG. 2 is a diagram showing a configuration example of the information providing device according to the present embodiment. FIG. 3 is a diagram showing an example of information stored in the question and answer storage unit according to the present embodiment. FIG. 4 is a flowchart showing an example of the flow of processing according to this embodiment. FIG. 5 is a diagram showing an example of the hardware configuration.

DESCRIPTION OF THE PREFERRED EMBODIMENTS An information providing apparatus, an information providing method, and an embodiment of an information providing program (hereinafter referred to as "embodiments") according to the present application will be described in detail below with reference to the drawings. Note that this embodiment does not limit the information providing apparatus, information providing method, and information providing program according to the present application.

[Embodiment]
[1. About the overview of the information providing device]
First, an example of the information providing apparatus 10 according to the present embodiment will be described using FIG. 1. In the following description, as an example of the process executed by the information providing device 10, a FAQ (question and answer collection) regarding a predetermined category specified by the user is generated using question and answer information accumulated on a so-called Q&A site. Describe the automatic creation process. This type of Q&A site uses the Internet to share knowledge and wisdom among users by having other users (respondents) post answers to questions posted by users (questioners). It is something to do.

Furthermore, it is assumed that the question text and the answer text in this embodiment are text data containing one or more sentences. It is also assumed that a sentence is a range of text data separated by periods, exclamation marks, question marks, spaces, etc. Additionally, a question text is a document that is processed as having been created with the intention of asking a question, and an answer text is a document that is processed as having been created with the intention of answering a question text. It is. Furthermore, in the following description, a user who instructs the creation of a FAQ using the information providing device 10 will be referred to as a user. Further, information received from each user, such as a question text and a corresponding answer text, may be collectively referred to as question and answer information or question and answer.

FIG. 1 is a diagram showing an example of an information providing apparatus according to this embodiment. The information providing device 10 shown in FIG. 1 is an information processing device that performs a process of creating FAQ, which is a collection of questions and answers, and is realized by, for example, a server device, a cloud system, or the like. For example, the information providing device 10 is a network such as various wireless communication networks such as 4G (Generation), 5G, LTE (Long Term Evolution), Wifi (registered trademark), or wireless LAN (Local Area Network), or various wired communication networks. Communication is performed between the terminal device 100 used by the user and the web server 200 that manages the Q&A site via N (for example, see FIG. 2).

The terminal device 100 is realized by a smart device such as a PC (Personal Computer), a server device, a smart television, a smartphone, or a tablet, and can communicate with the information providing device 10 via the network N. It is a mobile terminal device. In addition, the terminal device 100 has a screen such as a liquid crystal display, which has a touch panel function, and the information providing device 10 allows the user to perform tap operations, slide operations, scroll operations, etc. using a finger or a stylus, etc. It may have a function that can accept various operations on the content that is displayed.

Note that in the example shown in FIG. 1, the terminal device 100 is a terminal device used by the user U. Further, in the example shown in FIG. 1, one user U is described, but the present invention is not limited to this. Since any number of users can connect to the information providing device 10 via their respective terminal devices, each user can create the FAQ that he/she intends.

The web server 200 is an information processing device that manages the Q&A site, and is realized by, for example, a server device, a cloud system, or the like. The web server 200 receives questions and answers posted by users, and stores them in association with the answers corresponding to the questions. The web server 200 also provides the user with questions and answers searched on the Q&A site. In this embodiment, the web server 200 has been described as being separate from the information providing device 10, but they may be configured as one.

[2. Regarding an example of processing]
Since the Q&A site managed by the web server 200 has accumulated a large number of questions and answers, there is a desire for a service that uses these questions and answers to create and provide FAQs in various categories. In this case, it is preferable to automatically create FAQs because creating FAQs for various categories requires a huge amount of manpower. On the other hand, when trying to automatically create FAQs corresponding to various categories, the granularity and axes (direction) of question answers (headings) are not aligned, resulting in the problem of not being able to provide FAQs that match the user's intentions. There is a risk that this may occur. For example, when creating a FAQ related to credit cards, if questions and answers about each card company are mixed, the granularity of the questions and answers will vary. For this reason, it is preferable to reflect the user's intention to separate question and answers for each card company. Additionally, the questions and answers regarding payment methods (smartphone payments and e-commerce) are different from those regarding payment processing companies. For this reason, it is preferable to reflect the user's intention that they do not want questions and answers to be divided by payment method and by payment agency company at the same time.

In the present embodiment, the information providing device 10 performs distance learning on question answers that have been vectorized in advance, and then performs clustering processing, thereby suppressing variations in the granularity and axes of the question answers and providing user feedback. The aim is to provide FAQs that match the intent of the website. An example of each process executed by the information providing device 10 will be described below.

[2-1. About an example of preprocessing for converting to vector]
As shown in FIG. 1, the information providing device 10 receives a question and answer from the web server 200 (step S1). This question and answer may be sent by the web server 200 in response to an instruction from the information providing device 10, or may be sent periodically by the web server 200. Further, the web server 200 may transmit all questions and answers each time, or may transmit difference information from the previous time.

When the information providing device 10 receives the question and answer, it performs preprocessing to convert the question and answer into vector data (step S2). In the present embodiment, the information providing apparatus 10 generates an N-dimensional vector from each piece of text information using, for example, a learning model (autoencoder) that extracts feature amounts of text information of question and answers.

The learning model is configured to include, for example, an input layer, a middle layer, and an output layer. The input layer is a layer to which information is input, and the output layer is a layer to which information similar to the input information is output in response to input to the input layer. In this configuration, the section from the input layer to the middle layer corresponds to the part that compresses the input information (encoding processing), and the section from the middle layer to the output layer corresponds to the part that performs the processing to compress the compressed information. Corresponds to the part that performs processing (decoding processing). Furthermore, the intermediate layer is a layer that expresses the characteristics of information compressed between the input layer and the intermediate layer.

For example, the information providing device 10 calculates the value of each element (neuron) of the learning model M by inputting question and answer text information into the input layer of a predetermined learning model M, and calculates the value of each element (neuron) of the learning model M. Output information from the output layer. In this case, the information providing device 10 extracts, for example, the value of each element (neuron) of the intermediate layer as a feature amount, and generates N-dimensional vector data corresponding to the question answer. This vector data is expressed, for example, as an N-dimensional real number sequence.

Such a learning model M can be realized by, for example, various classifiers such as DNN (Deep Neural Network). Note that a neural network having an arbitrary configuration such as an RNN (Recurrent Neural Network), a CNN (Convolution Neural Network), or an LSTM (Long short-term memory) can be employed as the DNN.

Subsequently, when the user U specifies a category (for example, credit card) through the terminal device 100 (step S3), the information providing device 10 extracts vector data corresponding to question answers related to the field of credit cards ( Step S4). The information providing device 10 can extract vector data corresponding to question answers including vocabulary such as card company name, membership, withdrawal, billing, details, etc., for example. Thereby, the information providing device 10 can compose a question and answer that is mapped to a semantic space related to a predetermined category using the extracted vector data. Note that in this embodiment, all the accumulated question answers are vectorized by the learning model M, and then vector data related to the specified category is extracted, but the present invention is not limited to this. For example, questions and answers related to a specified category may be extracted from all accumulated question and answers, and the extracted questions and answers may be vectorized by the learning model M.

[2-2. Regarding an example of distance learning processing]

Subsequently, the user U transmits, through the terminal device 100, an instruction that includes the user's U's intention with respect to a plurality of question answers mapped to a semantic space related to a predetermined category (step S5). When this instruction is received, the information providing device 10 performs distance learning of a plurality of question answers mapped to a semantic space related to a predetermined category based on this instruction (step S6). This instruction is a small example for suppressing variations in the granularity and axes of multiple question answers, for example.

When creating an FAQ, for example, in an FAQ related to finance, it is okay to include questions and answers about each credit card company; If the answers are mixed, the granularity of the questions and answers will vary. Additionally, the questions and answers regarding payment methods (smartphone payments and e-commerce) are different from those regarding payment processing companies. Therefore, in order to provide an FAQ that is easy for users to use, it is effective to suppress variations in the granularity and axes of question answers. For this reason, the information providing device 10 executes distance learning in order to suppress variations in the granularity and axes of question and answers, and calculates the distance of vector data of question and answers in the semantic space (for example, relative distance in the semantic space). I'm making adjustments.

In this distance learning, similar data is learned to be relatively close in the semantic space, and dissimilar data is learned to be relatively far away in the semantic space. The information providing device 10 can perform Mahalanobis distance learning, for example. When executing this Mahalanobis distance learning, as an example of the user's instructions (teacher data), a set of credit card related data (similar data) regarding company A and card companies other than company A (company B, company C) are used. ...) and a set of credit card-related data (dissimilar data). The information providing device 10 learns a predetermined covariance matrix using these sets of similar data and sets of dissimilar data. Then, the information providing device 10 calculates the vector data of the question and answer using the learned covariance matrix, so that the credit card related data regarding company A is relatively closer to the credit card related data regarding credit card companies other than company A. data is relatively far away. Therefore, the distance between Company A's credit card-related question answers and credit card-related question answers of companies other than A can be increased. In this embodiment, an example of distance learning has been described, but the present invention is not limited to this.

[2-3. About an example of clustering processing]
Subsequently, the information providing device 10 performs clustering processing based on the distance learned vector data (step S7). This clustering process takes into consideration the distance (for example, cosine distance) between vector data, and forms hierarchical clusters based on question and answer groups corresponding to vector data with close distances. In this embodiment, the distance of vector data of question and answers is adjusted by distance learning before clustering processing, so clustering is performed with variations in the granularity and axes of the question and answers suppressed. Subsequently, the information providing device 10 creates a FAQ for each cluster formed by creating headings for questions and answers included in the cluster (step S8). The information providing device 10 analyzes the words of the questions and answers included in the cluster and uses the characteristic words to generate headings (for example, [Company A membership/withdrawal], [Company A details], [Company A points], etc.). Attach. Note that the central question and answer of each cluster may be selected instead of the heading. Finally, the information providing device 10 sends the created FAQ information to the terminal device 100 to provide the user with the FAQ (question and answer collection) (step S9). In this configuration, multiple question answers mapped to a semantic space related to a predetermined category are distance learned based on the user's instructions, so variations in the granularity and axes of question answers can be suppressed, and the user's intentions You can create and provide FAQs based on the following.

[3. Configuration of information providing device]
An example of the functional configuration of the information providing device 10 described above will be described below. In the following description, an example of a functional configuration of an information providing apparatus 10 that uses question and answer information accumulated in a Q&A site to create and provide a question and answer collection related to a category desired by a user will be described. FIG. 2 is a diagram showing a configuration example of the information providing device according to the present embodiment. As shown in FIG. 2, the information providing device 10 includes a communication section 20, a storage section 30, and a control section 40.

The communication unit 20 is realized by, for example, a NIC (Network Interface Card). The communication unit 20 is connected to the network N by wire or wirelessly, and transmits and receives information between the terminal device 100 and the web server 200.

The storage unit 30 is realized by, for example, a semiconductor memory element such as a RAM (Random Access Memory) or a flash memory, or a storage device such as a hard disk or an optical disk. Furthermore, the storage unit 30 includes a question and answer storage unit 31 and a learning model storage unit 32.

FIG. 3 is a diagram showing an example of information stored in the question and answer storage unit according to the present embodiment. The question and answer storage unit 31 stores various information regarding questions and answers. For example, the question and answer storage unit 31 stores question and answer IDs and vector data. In the example of FIG. 3, the question and answer storage unit 31 includes information on items such as "question and answer ID," "question and answer," "question and answer information," and "vector information."

"Question and answer ID" indicates identification information for identifying a question and answer. This question and answer consists of a question document and a corresponding answer document. Further, "Question and Answer" indicates the specific name, content, etc. of the question and answer identified by the question and answer ID. In the example of FIG. 3, "Question and Answer" is indicated by character information for specifying the content, such as "A company card membership".

"Question and answer information" indicates information regarding the question and answer identified by the question and answer ID. In the example in Figure 3, the question and answer information is ``Are there any documents required when applying for a Company A card?'' and ``In principle, there are no documents required when applying. In some cases, identification documents, etc. may be required.'' "Vector information" indicates vector information corresponding to the question and answer information identified by the question and answer ID. In the example of FIG. 3, the question and answer information corresponding to the question and answer ID is shown as multidimensional (N-dimensional) vector information (vector data) "10, 24, 54, 2, . . .".

The learning model storage unit 32 stores a learning model M that converts text information of question and answers into vector data (vectorization). This type of learning model M can be realized by various classifiers such as DNN (Deep Neural Network), for example. Note that a neural network having an arbitrary configuration such as an RNN (Recurrent Neural Network), a CNN (Convolution Neural Network), or an LSTM (Long short-term memory) can be employed as the DNN.

Returning to FIG. 2, the explanation will be continued. The control unit 40 is, for example, a controller, and various programs stored in a storage device inside the information providing device 10 use a RAM as a work area using a CPU (Central Processing Unit), an MPU (Micro Processing Unit), or the like. This is achieved by executing as . Further, the control unit 40 is a controller, and is realized by, for example, an integrated circuit such as an ASIC (Application Specific Integrated Circuit) or an FPGA (Field Programmable Gate Array).

As shown in FIG. 2, the control unit 40 includes a preprocessing unit 41, an extraction unit 42, a distance learning unit 43, a clustering processing unit 44, and an FAQ creation unit 45. The preprocessing unit 41 uses the learning model M stored in the learning model storage unit 32 to generate N-dimensional vector data from the text information of the question and answer.

For example, the preprocessing unit 41 calculates the value of each element (neuron) of the learning model M by inputting text information of a question answer to the input layer of the learning model M, and calculates information similar to the input text information. Output from the output layer. In this case, the information providing device 10 extracts, for example, the value of each element (neuron) of the intermediate layer as a feature amount, and generates N-dimensional vector data corresponding to the question answer. This vector data is expressed, for example, as an N-dimensional real number sequence, and is stored in the above-mentioned question and answer storage section 31 as "vector information" corresponding to the question and answer ID.

The extraction unit 42 extracts the corresponding vector data according to the user's instructions. For example, when the extraction unit 42 specifies credit cards as a category, it extracts vector data corresponding to questions and answers related to the field of credit cards. In this case, the extraction unit 42 can extract vector data corresponding to the question and answer including vocabulary such as card company name, membership, withdrawal, billing, statement, etc., as information related to the field of credit cards. Thereby, the information providing device 10 can use the extracted vector data to compose a question and answer mapped to a semantic space regarding a predetermined category (credit card).

The distance learning unit 43 executes distance learning to adjust the distance of vector data of question and answers mapped to a semantic space related to a predetermined category in response to a user's instruction. The user's instructions are, for example, examples (teacher data) for suppressing variations in the granularity and axes of question answers. When creating an FAQ, for example, in an FAQ related to finance, it is okay to include questions and answers about each credit card company; If the answers are mixed, the granularity of the questions and answers will vary. Additionally, the questions and answers regarding payment methods (smartphone payments and e-commerce) are different from those regarding payment processing companies. Therefore, in order to provide an FAQ that is easy for users to use, it is effective to suppress variations in the granularity and axes of question answers. For this reason, the distance learning unit 43 performs distance learning on the question answers mapped to the semantic space, and adjusts the distances (for example, relative distances in the semantic space) of the vector data of these question answers. The granularity of answers and variations in axes are suppressed.

In this distance learning, similar data is learned to be relatively close in the semantic space, and dissimilar data is learned to be relatively far away in the semantic space. The information providing device 10 can perform Mahalanobis distance learning, for example. When executing this Mahalanobis distance learning, as an example of the user's instructions (teacher data), a set of credit card related data (similar data) regarding company A and card companies other than company A (company B, company C) are used. ...) and a set of credit card-related data (dissimilar data). The distance learning unit 43 learns a predetermined covariance matrix using these sets of similar data and sets of dissimilar data. Then, the distance learning unit 43 calculates the vector data of the question and answer using the learned covariance matrix, so that the credit card related data related to company A are relatively close, and the credit card related data related to card companies other than company A are calculated. data is relatively far away. Therefore, the distance between Company A's credit card-related question answers and credit card-related question answers of companies other than A can be increased.

The clustering processing unit 44 executes clustering processing based on the distance learned vector data. The clustering processing unit 44 takes into consideration the distance between vector data (for example, cosine distance) and forms hierarchical clusters using question and answer groups corresponding to vector data with a close distance. For example, if a hierarchical structure is required, the Ward method or the group average method can be used as the clustering method. Furthermore, if a non-hierarchical structure is required, the k-means method can be used. Further, the distance used in the clustering process is not limited to the cosine distance, and for example, an existing distance such as the Euclidean distance may be used.

In this embodiment, before the clustering processing unit 44 executes the clustering process, the distance learning unit 43 performs distance learning to adjust the distance of the vector data of the question and answer, so the granularity and axis of the question and answer are adjusted. Clustering processing can be performed while suppressing the variation in .

The FAQ creation unit 45 creates FAQs by creating headings for questions and answers included in each cluster formed by the clustering process. The FAQ creation unit 45 analyzes the words of the questions and answers included in the cluster and creates headings (for example, [Company A membership/withdrawal], [Company A details], [Company A points], etc.) using characteristic words. Attach. In this case, the FAQ creation unit 45 extracts words (nouns and noun suffixes) included in the text information of the questions and answers of the cluster, and extracts features based on the frequencies of these words (nouns and noun suffixes). Note that the FAQ creation unit 45 may select the central question and answer of each cluster instead of the heading. In this case, the FAQ creation unit 45 selects the question answer that is closest in distance from the center point of each cluster (has the highest degree of similarity) as the central question answer.

The FAQ creation unit 45 sends information regarding the created FAQ to the terminal device 100 via the communication unit 20, thereby providing the user with an FAQ (question and answer collection). The provision of FAQs may be performed by a part other than the FAQ creation section 45, or, for example, a provision section may be provided separately and the provision section may execute the provision. In this embodiment, the distance learning unit 43 performs distance learning on a plurality of question answers mapped to a semantic space related to a predetermined category based on instructions from the user, so it is possible to suppress variations in the granularity and axes of question answers. The FAQ creation unit 45 can create and provide FAQs that meet the user's intentions.

[4. Processing procedure]
Next, an example of the flow of processing executed by the information providing apparatus 10 according to this embodiment will be described using FIG. 4. FIG. 4 is a flowchart showing an example of the flow of processing according to this embodiment. As shown in FIG. 4, the information providing apparatus 10 performs preprocessing to convert the question and answer received from the web server 200 into vector data (step S101). Subsequently, the information providing device 10 extracts vector data corresponding to the question and answer related to the category based on the user's instruction (step S102). Subsequently, the information providing device 10 executes distance learning of a plurality of question answers mapped to the semantic space related to a predetermined category based on the user's instructions (step S103). Subsequently, the information providing apparatus 10 performs clustering based on the distance learned vector data (step S104). Next, for each cluster formed by clustering, the information providing device 10 creates a FAQ by creating a heading for the questions and answers included in the cluster (step S105), and provides this FAQ to the user. Finish the process. Note that the information providing apparatus 10 may change the order in which steps S101 and S102 are executed. Further, step S101, step S102, and step S103 may be executed simultaneously.

[5. Modified example]
The information providing device 10 described above may be implemented in various different forms other than the embodiments described above. Therefore, other embodiments of the information providing device 10 will be described below.

[5-1. Regarding simultaneous processing of preprocessing and distance learning processing]
In the above-described embodiment, the information providing device 10 performs distance learning processing to perform distance learning of these question answers after preprocessing to map a plurality of question answers to a semantic space related to a predetermined category. It is also possible to process the processing and distance learning processing almost simultaneously to perform learning at once.

For example, the information providing device 10 uses the terminal device 100 to specify categories (search queries and existing question sets) specified by the user, and specific examples of classification specified by the user (labels that the user adds based on the search results). ) to obtain. The information providing device 10 generates an N-dimensional vector from the text information of each question and answer using one of the learning models (autoencoder) for a plurality of questions and answers regarding the specified category. Further, the information providing device 10 can adjust the vectorized question-answer distance using another learning model generated by a similarity learning method such as triplet loss, for example. In this case, it is possible to add triplet losses such that examples that you want to have in the same cluster are close together, and examples that you want to have in different clusters are far away. You can also add a triplet loss so that questions in the same category are close together and others are far apart.

[5-2. others〕
In addition, as a clustering process, the vector data of the obtained question answers is converted into a graph using a method such as the ε-nearest neighbor method, k-nearest neighbor method, or fully connected method, and then clustering is performed by focusing on the connectivity of this graph. Spectral clustering can also be used. During this clustering, since clustering is performed after reducing the expression to a low-dimensional representation that is easy to cluster, the weights of the question and answer axes can be adjusted to fit the training data specified by the user.

Furthermore, in the embodiment described above, the similarity of texts is measured based on the distance between texts (for example, cosine distance), but for example, the similarity of texts can be measured using a concept graph such as WordNet. Good too. WordNet is a so-called concept dictionary, and since word IDs and concept IDs are linked, it is possible to measure the degree of similarity between words of higher-level and lower-level concepts that have similar concepts, even if the words are dissimilar. For example, the information providing device 10 can measure the similarity of texts by using a learning model that learns vectors that preserve a hierarchical structure on Poincaré space.

[6. effect〕
As described above, the information providing device 10 includes a distance learning unit 43 that performs distance learning based on user instructions on a plurality of accumulated question and answer documents, and clusters the distance learned question and answer documents. Since it includes a clustering processing unit 44 and an FAQ creation unit 45 that summarizes each cluster including clustered question and answer documents and creates an FAQ, it is possible to suppress variations in the granularity and axes of question and answers, and to You can create FAQs that match your intentions.

In addition, since it includes a preprocessing unit 41 that performs preprocessing to vectorize a plurality of question and answer documents and map them to a predetermined semantic space before performing distance learning, the question and answer documents mapped to the semantic space are , it is possible to easily perform distance learning based on user instructions.

In addition, the distance learning unit 43 moves the question-and-answer document within the semantic space by calculating the vectorized question-and-answer document using the covariance matrix learned based on the user's instructions. Variations in the granularity and axes of question answers can be easily suppressed.

In addition, the clustering processing unit 44 takes into account the distance between a plurality of vectorized question-and-answer documents and forms a cluster by a group of question-and-answers corresponding to vectors with a close distance. Clustering can be performed with reduced variation.

Furthermore, since the FAQ creation unit 45 attaches headings to the question and answer documents included in each cluster, it is possible to easily create the FAQ.

Moreover, since the preprocessing unit 41 and the distance learning unit 43 simultaneously perform preprocessing and distance learning, respectively, they can quickly process distance learning based on a user's instruction.

[7. Hardware configuration]
Further, the information providing apparatus 10 according to the embodiments described above is realized by, for example, a computer 1000 having a configuration as shown in FIG. FIG. 5 is a hardware configuration diagram showing an example of a computer that implements the functions of the information providing device 10 according to the embodiment. Computer 1000 has CPU 1100, RAM 1200, ROM 1300, HDD 1400, communication interface (I/F) 1500, input/output interface (I/F) 1600, and media interface (I/F) 1700.

CPU 1100 operates based on a program stored in ROM 1300 or HDD 1400, and controls each section. The ROM 1300 stores a boot program executed by the CPU 1100 when the computer 1000 is started, programs depending on the hardware of the computer 1000, and the like.

The HDD 1400 stores programs executed by the CPU 1100, data used by the programs, and the like. Communication interface 1500 receives data from other devices via network (communication network) N and sends it to CPU 1100, and sends data generated by CPU 1100 to other devices via network N.

The CPU 1100 controls output devices such as a display and a printer, and input devices such as a keyboard and mouse via an input/output interface 1600. CPU 1100 obtains data from an input device via input/output interface 1600. Further, CPU 1100 outputs the generated data to an output device via input/output interface 1600.

Media interface 1700 reads programs or data stored in recording medium 1800 and provides them to CPU 1100 via RAM 1200. CPU 1100 loads this program from recording medium 1800 onto RAM 1200 via media interface 1700, and executes the loaded program. The recording medium 1800 is, for example, an optical recording medium such as 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. etc.

For example, when the computer 1000 functions as the information providing apparatus 10 according to the embodiment, the CPU 1100 of the computer 1000 realizes the functions of the control unit 40 by executing a program loaded onto the RAM 1200. The CPU 1100 of the computer 1000 reads these programs from the recording medium 1800 and executes them, but as another example, these programs may be acquired from another device via the network N.

Some of the embodiments of the present application have been described above in detail based on the drawings, but these are merely examples, and various modifications and variations may be made based on the knowledge of those skilled in the art, including the embodiments described in the disclosure section of the invention. It is possible to carry out the invention in other forms with modifications.

[8. others〕
Furthermore, among the processes described in the above embodiments and modified examples, all or part of the processes described as being performed automatically can be performed manually, or may be described as being performed manually. All or part of this processing can also be performed automatically using known methods. In addition, information including the processing procedures, specific names, and various data and parameters shown in the above documents and drawings may be changed arbitrarily, unless otherwise specified. For example, the various information shown in each figure is not limited to the illustrated information.

Furthermore, each component of each device shown in the drawings is functionally conceptual, and does not necessarily need to be physically configured as shown in the drawings. In other words, the specific form of distributing and integrating each device is not limited to what is shown in the diagram, and all or part of the devices can be functionally or physically distributed or integrated in arbitrary units depending on various loads and usage conditions. Can be integrated and configured. For example, the preprocessing unit 41, the extraction unit 42, the distance learning unit 43, the clustering processing unit 44, or the FAQ creation unit 45 may be connected as external devices to the information providing device 10 via a network. In addition, separate devices each have a preprocessing section 41, an extraction section 42, a distance learning section 43, a clustering processing section 44, or an FAQ creation section 45, and by being connected to a network and working together, the above-mentioned information providing device 10 functions may be realized. In other words, the information providing device 10 may be realized by a plurality of server computers, and depending on the function, it may be realized by calling an external platform etc. using an API (Application Programming Interface) or network computing, etc., so the configuration is flexible. can be changed to

Furthermore, the above-described embodiments and modifications can be combined as appropriate within a range that does not conflict with the processing contents.

Further, the above-mentioned "section, module, unit" can be read as "means", "circuit", etc. For example, the preprocessing section 41 can be read as a preprocessing means or a preprocessing circuit.

10 information providing device 20 communication unit 30 storage unit 31 question and answer storage unit 32 learning model storage unit 40 control unit 41 preprocessing unit 42 extraction unit 43 distance learning unit (learning processing unit)
44 Clustering processing unit 45 FAQ creation unit 100 Terminal device

Claims (6)

a learning processing unit that performs distance learning for adjusting the relative distance of vector data of question and answers mapped to a semantic space related to a predetermined category specified by a user with respect to a plurality of accumulated question and answer documents;
a clustering processing unit that clusters the learned question and answer documents;
an FAQ creation unit that creates a question and answer collection using the question and answer closest to the center point of each cluster including the clustered question and answer documents;
An information providing device comprising: The learning processing unit calculates the vectorized question and answer document using a covariance matrix that has undergone predetermined learning using a set of similar data and a set of dissimilar data for the vector data. , the information providing apparatus according to claim 1, wherein the relative distance of the vector data of the question and answer is adjusted within the semantic space. 2. The clustering processing unit takes into account the distance between the plurality of vectorized question and answer documents, and forms a cluster by a group of question and answers corresponding to vector data having a close distance. 2. The information providing device according to 2. 4. The information providing apparatus according to claim 1, wherein the FAQ creation unit attaches a heading to the question and answer documents included in each cluster. An information provision method executed by a computer, the method comprising:
a learning processing step of performing distance learning for adjusting the relative distance of vector data of question and answers mapped to a semantic space related to a predetermined category specified by a user with respect to the plurality of accumulated question and answer documents;
a clustering process for clustering the learned question and answer documents;
an FAQ creation step of creating a question and answer collection using the question and answer closest to the center point of each cluster including the clustered question and answer documents;
An information provision method characterized by including. a learning processing procedure for performing distance learning for adjusting the relative distance of vector data of question and answers mapped to a semantic space related to a predetermined category specified by a user for a plurality of accumulated question and answer documents;
a clustering procedure for clustering learned question-and-answer documents;
A FAQ creation procedure for creating a question and answer collection using the question and answer closest to the center point of each cluster including clustered question and answer documents;
An information providing program that causes a computer to execute.

Images