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

Abstract

An information processing device (1), in order to enable the presentation of useful information when recommending a recommendation, comprises: a generation unit (11) that generates questions or hypotheses according to the recommendation; and a presentation unit (12) that presents answers to the questions generated by the generation unit (11) or the results of verifying the hypotheses generated by the generation unit (11).

Description

Technical Field

[0001] It relates to a technology for recommending products and services.

Background Art

[0002] Technologies for recommending products and services have been conventionally known. For example, in Patent Document 1 below, based on purchase history information of products and product factors indicating reasons for purchasing products, a customer group by purchase reason tendency is generated, and a product recommendation device that determines recommended products to be recommended to a customer according to which customer group by purchase reason tendency the customer belongs to is disclosed.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] The product recommendation device described in Patent Document 1 can only determine recommended products, and cannot present useful information when recommending the determined recommended products, and there is room for improvement in this regard. One aspect of the present invention has been made in view of the above viewpoints, and an example of its object is to provide an information processing device or the like that can present useful information when recommending a recommendation target.

Means for Solving the Problems

[0005] An information processing device according to one aspect of the present invention includes a generation means for generating a question or hypothesis corresponding to a recommendation target determined as something to be recommended to a target person, and a presentation means for presenting an answer to the question generated by the generation means or a verification result of the hypothesis generated by the generation means.

[0006] An information presentation method relating to one aspect of the present invention includes, in which at least one processor generates questions or hypotheses corresponding to recommendation targets determined to be recommended to a target person, and presents answers to the generated questions or the results of verifying the generated hypotheses.

[0007] An information presentation program according to one aspect of the present invention causes a computer to function as a generation means for generating questions or hypotheses corresponding to recommendation targets determined to be recommended to a target person, and as a presentation means for presenting answers to the questions generated by the generation means, or the results of verifying the hypotheses generated by the generation means. [Effects of the Invention]

[0008] According to one aspect of the present invention, it becomes possible to present useful information when recommending a subject for recommendation. [Brief explanation of the drawing]

[0009] [Figure 1] This is a block diagram showing the configuration of an information processing device according to an exemplary embodiment 1 of the present invention. [Figure 2] This is a flowchart showing the flow of the information presentation method according to Exemplary Embodiment 1 of the present invention. [Figure 3] This figure shows an overview of the information presentation method according to Exemplary Embodiment 2 of the present invention. [Figure 4] This is a block diagram showing the configuration of an information processing device according to an exemplary embodiment 2 of the present invention. [Figure 5] This figure shows examples of generating a predictive model and generating recommendation reasons. [Figure 6] This is a flowchart showing the flow of the information presentation method according to exemplary embodiment 2 of the present invention. [Figure 7] This figure shows an example of a computer that executes program instructions, which are software that realizes each function of each exemplary embodiment of the present invention. [Modes for carrying out the invention]

[0010] [Exemplary Embodiment 1] A first exemplary embodiment of the present invention will be described in detail with reference to the drawings. This exemplary embodiment is the basic form of the exemplary embodiments described later.

[0011] (Configuration of Information Processing Device 1) The configuration of the information processing device 1 according to this exemplary embodiment will be described with reference to Figure 1. Figure 1 is a block diagram showing the configuration of the information processing device 1. As shown in Figure 1, the information processing device 1 includes a generation unit (generation means) 11 and a presentation unit (presentation means) 12.

[0012] The generation unit 11 generates questions or hypotheses corresponding to the recommendation target that has been determined to be recommended to the target person. The generation unit 11 may generate both questions and hypotheses.

[0013] The presentation unit 12 presents the answer to the question generated by the generation unit 11, or the verification result of the hypothesis generated by the generation unit 11. If the generation unit 11 generates both a question and a hypothesis, the presentation unit 12 may present both the answer to the question and the verification result of the hypothesis. Furthermore, the generation of the answer to the question and the verification of the hypothesis may be performed by the information processing device 1, or may be performed by another information processing device.

[0014] As described above, the information processing device 1 according to this exemplary embodiment includes a generation unit 11 that generates questions or hypotheses corresponding to recommendation targets determined to be recommended to the target person, and a presentation unit 12 that presents answers to the questions generated by the generation unit 11, or the results of verifying the hypotheses generated by the generation unit 11. Therefore, according to the information processing device 1 according to this exemplary embodiment, it is possible to present useful information when recommending recommendation targets.

[0015] (Information presentation program) The functions of the above-described information processing apparatus 1 can also be realized by a program. The information presentation program according to this exemplary embodiment causes a computer to function as the generation unit 11 and the presentation unit 12. According to this information presentation program, an effect that it becomes possible to present information useful when recommending a recommendation target can be obtained.

[0016] (Flow of the information presentation method) The flow of the information presentation method according to this exemplary embodiment will be described with reference to FIG. 2. FIG. 2 is a flowchart showing the flow of the information presentation method. In addition, the execution subject of each step in this information presentation method may be a processor included in the information processing apparatus 1, may be a processor included in another apparatus, or may be processors provided in different apparatuses for each step.

[0017] In S11, at least one processor generates a question or hypothesis corresponding to the recommendation target determined as something to be recommended to the target person. In addition, in S11, both a question and a hypothesis may be generated.

[0018] In S12, at least one processor presents an answer to the question generated in S11 or a verification result of the hypothesis generated in S11. In addition, when both a question and a hypothesis are generated in S11, in S12, both an answer to the question and a verification result of the hypothesis may be presented.

[0019] As described above, the information presentation method according to this exemplary embodiment includes at least one processor generating a question or hypothesis corresponding to the recommendation target determined as something to be recommended to the target person, and presenting an answer to the generated question or a verification result of the generated hypothesis. According to this information presentation method, an effect that it becomes possible to present information useful when recommending a recommendation target can be obtained.

[0020] 〔Exemplary Embodiment 2〕 (Outline of the information presentation method) Figure 3 is a diagram illustrating an overview of the information presentation method (hereinafter referred to as "this method") according to this exemplary embodiment. Figure 3 shows two people, Person A and Person B. Person B is a recommender who recommends a product or service, and Person A is the person being recommended.

[0021] Person B could be, for example, a salesperson. In this case, how and what products Person B recommends to Person A will affect the success or failure of the sale, that is, whether or not the recommended products are sold. This method makes it possible to make Person B's sales activities more effective by providing Person B with useful information when recommending products.

[0022] In this method, first, the person who is the subject of the recommendation... A Attribute data D1, which indicates what kind of person person A is, is input into the predictive model 211, and the probability of a sale being concluded when each candidate product is recommended to person A is predicted. In the example in Figure 3, of the candidate products A and B, the probability of a sale for product A is predicted to be 0.82, and the probability of a sale for product B is predicted to be 0.1. Note that the recommended products A and B may be physical goods, services, or a combination of both. Further details of the predictive model 211 will be described later.

[0023] In this example, the numerical range of the probability is set from 0 to 1, so it can be said that there is a high probability of closing a deal when recommending product A, and a low probability of closing a deal when recommending product B. Therefore, in this example, product A is determined to be the recommended product. Note that whether or not the probability of closing a deal is high can be determined based on a predetermined threshold. In other words, in this method, products whose probability obtained using the prediction model 211 is above the threshold may be selected as the recommended products.

[0024] Next, this method generates reasons for recommending the selected person. In the example in Figure 3, the reasons for recommending person A are that their hobby is golf and their annual income is over 7 million yen. These reasons are useful information for recommending product A. However, based solely on the reasons for recommending shown in Figure 3, person B cannot understand the relationship between person A's characteristics—that their hobby is golf and their annual income is over 7 million yen—and product A. Therefore, person B may not be able to effectively persuade person A to buy product A.

[0025] Therefore, this method generates questions and hypotheses corresponding to the target audience determined as described above. In the example in Figure 3, the questions "What is golf useful for?" and "What kind of person earns 7 million yen a year?" are generated, along with the hypothesis "Product A is related to golf." Note that this method may generate only questions or hypotheses. Also, one question or multiple questions may be generated. The same applies to hypotheses. The method for generating questions and hypotheses will be described later.

[0026] This method then generates responses to the questions and hypotheses generated as described above. In the example in Figure 3, the answer "Networking and health improvement" is generated to the question "What is golf useful for?", and the answer "A person in their prime as a business person" is generated to the question "What kind of person earns 7 million yen a year?". In addition, the verification result "Not directly related" is generated for the hypothesis "Product A is related to golf".

[0027] In this method, the responses and verification results generated as described above are presented to person B. This information is useful for person B when recommending product A to person A. For example, in the example in Figure 3, person B, upon receiving the responses and verification results, realizes that customers with an annual income of 7 million yen or more often play golf as part of their career development. Person B then comes up with the idea of ​​proposing a plan that allows for network building in combination with product A. In this way, the responses and verification results presented by this method are useful information that contributes to making person B's sales activities more effective.

[0028] As described above, this method includes generating questions or hypotheses corresponding to the product to be recommended to the subject (person A in the example of Figure 3) (product A in the example of Figure 3), and presenting the answers to the generated questions or the results of verifying the generated hypotheses. Therefore, this method has the effect of making it possible to present useful information when recommending the product.

[0029] Furthermore, the information presented in this method may be directed to the person being recommended. For example, some online shopping sites automatically determine and display recommended products to their visitors. In such shopping sites, this method may be used to generate questions or hypotheses related to the recommended products, and the answers or verification results may be presented along with the recommended products. This allows visitors to understand the reasons and background for the recommendation of the product, or to gain various information about the product itself, thereby motivating them to make a purchase.

[0030] (Configuration of Information Processing Device 2) The configuration of the information processing device 2 according to this exemplary embodiment will be described with reference to Figure 4. Figure 4 is a block diagram showing the configuration of the information processing device 2. As shown in the figure, the information processing device 2 includes a control unit 20 that controls all parts of the information processing device 2, and a storage unit 21 that stores various data used by the information processing device 2. The information processing device 2 also includes an input unit 22 that receives user input operations to the information processing device 2, an output unit 23 for the information processing device 2 to output data, and a communication unit 24 for the information processing device 2 to communicate with other devices.

[0031] Furthermore, the control unit 20 includes a recommendation unit (recommendation means) 201, a recommendation reason generation unit (recommendation reason generation means) 202, a generation unit (generation means) 203, a response unit (response means) 204, and a presentation unit (presentation means) 205. The storage unit 21 stores the prediction model 211 and the generation model 212.

[0032] The recommendation unit 201 determines what to recommend to the target person. As mentioned above, the recommended items may be goods, services, or a combination thereof. Furthermore, it is desirable that the recommended items be tailored to the attributes of the target person. For example, the recommendation unit 201 may use a predictive model 211 to determine recommended items that are tailored to the attributes of the target person. The predictive model 211 will be explained later in the section "Method for Determining Recommended Items and Generating Recommendation Reasons".

[0033] The recommendation reason generation unit 202 generates the recommendation reasons for the recommended target determined by the recommendation unit 201. More precisely, the recommendation reason generation unit 202 generates information indicating the recommendation reasons for the recommended target determined by the recommendation unit 201, but here we will simply refer to the information indicating the recommendation reasons as "recommendation reasons." The method for generating recommendation reasons will be explained later in the section "Method for Determining Recommended Targets and Method for Generating Recommendation Reasons."

[0034] The generation unit 203 generates questions or hypotheses corresponding to the recommendation target determined by the recommendation unit 201. More precisely, the generation unit 203 generates question sentences, which are sentences that express questions in natural language, or hypothesis sentences, which are sentences that express hypotheses in natural language. Here, however, hypothesis sentences are simply referred to as hypotheses, and question sentences are simply referred to as questions. The generative model 212 can also be used to generate questions or hypotheses. Note that the question generation unit that generates questions and the hypothesis generation unit that generates hypotheses may be provided as separate blocks. Details of the question and hypothesis generation method and the generative model 212 will be explained later in the section "Question and Hypothesis Generation Method".

[0035] The response unit 204 generates responses to the questions or hypotheses generated by the generation unit 203. Alternatively, a response generation unit that generates answers to questions and a hypothesis verification unit that generates hypothesis verification results may be provided as separate blocks. As mentioned above, since both the questions and hypotheses generated by the generation unit 203 are in text form, it is possible to generate responses to them using natural language processing techniques. Details of the methods for generating answers to questions and verifying hypotheses will be explained later in the section "Response Generation and Hypothesis Verification Methods".

[0036] The presentation unit 205 presents the answer to the question generated by the generation unit 203, or the verification result of the hypothesis generated by the generation unit 203. The method of presentation is acceptable as long as it allows the target to recognize the content to be presented. For example, if the output unit 23 is a display device, the presentation unit 205 may present the answer or verification result by displaying it on the output unit 23. Alternatively, if the output unit 23 is an audio output device, the presentation unit 205 may present the answer or verification result by outputting it as audio on the output unit 23. The presentation unit 205 may also have the answer or verification result output to an external device of the information processing device 2.

[0037] As described above, the information processing device 2 according to this exemplary embodiment includes a generation unit 203 that generates questions or hypotheses corresponding to recommendation targets determined to be recommended to the target person, and a presentation unit 205 that presents answers to the questions generated by the generation unit 203, or the verification results of the hypotheses generated by the generation unit 203. Therefore, according to the information processing device 2 according to this exemplary embodiment, it is possible to present useful information when recommending recommendation targets.

[0038] (Method for determining who to recommend and method for generating reasons for recommendation) The method by which the recommendation unit 201 determines the target of recommendation and the method by which the recommendation reason generation unit 202 generates the recommendation reasons will be explained with reference to Figure 5. Figure 5 shows an example of the generation of the prediction model 211 and an example of the generation of recommendation reasons.

[0039] In the example shown in Figure 5, a predictive model 211 is generated through training using the training data D2. The training data D2 consists of data for each of several customers, indicating their ID (identification), annual income and hobbies, the products offered to them for purchase, and whether or not a sale of those products was made.

[0040] The training data D2 shows the relationship between a person's attributes and the results of recommending products to people with those attributes. Therefore, by learning using the training data D2, it is possible to generate a predictive model 211 that predicts the probability of a sale being made when a specific product is recommended to a target person, based on the target person's attributes.

[0041] The prediction model 211 is information that represents the relationship between the explanatory variables and the target variable. The prediction model 211 is a component for estimating the result of the target variable by calculating the target variable based on the explanatory variables. The prediction model 211 is generated by running a learning algorithm with training data in which the value of the target variable is already obtained and arbitrary parameters as input. The prediction model 211 may be represented, for example, by a function c that maps the input x to the correct answer y. The prediction model 211 may also estimate the numerical value of the target variable, or it may estimate the label of the target variable. Furthermore, the prediction model 211 may output a variable that describes the probability distribution of the target variable. The prediction model 211 may also be referred to as a "learning model," "analysis model," "AI (Artificial Intelligence) model," "trained model," "inference model," or "prediction formula." Explanatory variables are variables used as input in the prediction model. Explanatory variables may also be referred to as "features" or "features."

[0042] The predictive model 211 only needs to be able to predict the probability of a deal being closed, and the learning algorithm for generating the predictive model is not particularly limited. For example, the learning algorithm for generating the predictive model 211 may be a random forest, a support vector machine, a naive Bayes, or a neural network.

[0043] Alternatively, the prediction model 211 may be a piecewise linear model. A piecewise linear model is constructed by setting up categories that allow prediction by a linear model, and generating a linear model for each category. For example, suppose there are three categories set up: Category 1 where the attribute value of "hobbies" is "golf" and the attribute value of "annual income" is "700" or more; Category 2 where the attribute value of "annual income" is "1200" or more; and Category 3 where the attribute value of "annual income" is "400" or less. In this case, for each of these categories, a linear model is generated that predicts the probability of closing a deal when product A is recommended, based on the attribute values ​​of the target person.

[0044] Furthermore, one method for generating piecewise linear models utilizes FAB inference (Factorized Asymptotic Bayesian Inference). Methods for generating piecewise linear models using FAB inference are disclosed, for example, in U.S. Patent Publication US2014 / 0222741A1.

[0045] When the recommendation unit 201 determines a target for recommendation using a piecewise linear model, it identifies the target's category based on their attributes and predicts the probability of closing a deal using the linear model for that category. For example, the person in attribute data D1 shown in Figure 5 (the person with customer ID 2011) has an annual income of 720 and a hobby of golf. Therefore, the recommendation unit 201 identifies that this person belongs to category 1 and uses the linear model corresponding to category 1 to predict the probability of closing a deal when recommending product A to this person, based on the attribute values ​​shown in attribute data D1 (for example, the value of "annual income").

[0046] As described above, the recommendation unit 201 can predict the probability that each of the recommended candidates will be sold using the prediction model 211. Based on these predictions, the recommendation unit 201 can then decide which candidates to recommend. For example, the recommendation unit 201 may decide which candidate has the highest probability of being sold, or it may decide which candidates have a probability of being sold that is above a predetermined threshold.

[0047] Next, the method for generating recommendation reasons will be explained. If the prediction model 211 is a piecewise linear model, the recommendation reason generation unit 202 may use the conditions defined in the piece to which the target person belongs as the recommendation reason. For example, a person with customer ID 2011, as shown in attribute data D1, belongs to piece 1 (the piece where hobby = golf and annual income > 700) among the pieces 1 to 3 of the piecewise linear model described above. Therefore, the recommendation reason generation unit 202 may use the fact that the person's hobby is golf and their annual income is 7 million yen or more as the recommendation reason for recommending product A to that person.

[0048] Graph G1, shown in Figure 5, shows the number of successful transactions for each of the three categories (1-3) of the aforementioned piecewise linear model. In Graph G1, categories with a high number of successful transactions are placed higher, and categories with a high number of unsuccessful transactions are placed lower. The customer with customer ID 2011, as shown in attribute data D1, belongs to category 1 (hobby = golf and annual income > 7 million) which has the highest number of successful transactions in Graph G1. Therefore, the fact that this person's hobby is golf and their annual income is 7 million yen or more is a valid reason (or justification) for recommending product A to this person.

[0049] The methods for determining recommended targets and generating recommendation reasons described above are merely examples. For instance, a predictive model generated by learning the relationship between a person's attributes and the types of recommendations that should be given to that person may be used to determine recommended targets. In this case, the output of the predictive model would be the recommended targets.

[0050] Furthermore, the format of the input data to be input to the information processing device 2 for determining the recommended items is not particularly limited. For example, attribute data D1 in a table format as shown in Figure 5 can be used as input data, or other data formats such as images and audio can also be used as input data. The input data can be converted in format as necessary before being used to determine the recommended items.

[0051] Furthermore, the prediction algorithm is not particularly limited; for example, it may use predefined rules, such as attribution models, to predict the probability of a sale being completed. It is also possible to determine the target of recommendations without using a prediction model generated by machine learning. For example, if purchase history information for a product or service is available, customers may be pre-classified into multiple groups based on their purchase trends and customer attributes using this purchase history information. In this case, the recommendation unit 201 may classify the target individuals into one of these groups and determine the target of recommendations based on the purchase trends of that group (for example, those with a high purchase frequency or a high total purchase amount within that group).

[0052] Furthermore, any method appropriate to the method used to determine the target of recommendation can be adopted as the method for generating the recommendation reason. For example, the recommendation reason generation unit 202 may use as the recommendation reason an attribute among the attributes of the target person used to determine the target of recommendation that has a relatively strong correlation with the closing of a sale of the product. For example, suppose that in past cases, a common attribute among many people who closed a sale of a particular product was that their hobby was watching videos. In this case, if the target person's attribute includes watching videos as a hobby, the recommendation reason generation unit 202 may use the fact that their hobby is watching videos, or that the closing rate for people whose hobby is watching videos is high for that product, as the recommendation reason for the product.

[0053] (Methods for generating questions and hypotheses) The method for generating questions and hypotheses by the generation unit 203 will now be described. Various methods can be applied to generate questions and hypotheses, but for example, the generation unit 203 may generate questions and hypotheses using a generation model 212 generated by learning the relationship between recommendation targets and questions or hypotheses corresponding to those recommendations. This provides the effect of generating reasonable questions or hypotheses based on learning results, in addition to the effects of the information processing device 1 according to the exemplary embodiment 1.

[0054] For example, by training with training data that associates various information about the target of recommendation with questions that sales representatives have based on that information, and hypotheses that sales representatives have conceived based on that information, a generative model 212 that generates questions and hypotheses from various information about the target of recommendation can be generated. Note that the generative model is not limited to one trained using training data. The generative model may also be an unsupervised model such as a GAN (Generative Adversarial Network).

[0055] Various types of information regarding the recommended product include, for example, the attributes of the product (e.g., product name, product category, price or price range, target age, etc.), the reason for the recommendation, and the attributes of the target person or recommender (e.g., age, gender, occupation, income, career, affiliation). In addition, other information regarding the product may include, for example, the probability of a sale being made for the product (as predicted by prediction model 211).

[0056] By using the attributes of the target audience as information about the recommended candidates, it is possible to generate questions or hypotheses tailored to the target audience. For example, it is possible to generate questions or hypotheses tailored to the gender and age group of the target audience, thereby providing answers and hypothesis testing results tailored to the gender and age group of the target audience.

[0057] Furthermore, by using the attributes of the recommender as information about the person being recommended, it is possible to generate questions or hypotheses tailored to the recommender. For example, it is possible to generate questions or hypotheses based on the number of years the recommender has worked as a salesperson, and then present the answers and the results of hypothesis testing. This allows the recommender, upon receiving the information, to provide explanations to the person being recommended that are appropriate to their length of service.

[0058] Furthermore, by using information about the recommended product, such as the probability of the recommended product being sold or the probability that the recommended product is suitable for the target customer, it is possible to generate questions or hypotheses corresponding to the probability and present answers to those questions or the results of verifying those hypotheses. For example, if the probability is above a predetermined threshold, the generation unit 203 may generate questions or hypotheses that include predetermined wording that reflects the high probability (e.g., we can confidently recommend this product, we especially recommend this product, it is ideal for the target customer, etc.). Alternatively, for example, the generation unit 203 may generate questions or hypotheses using different generation models or templates depending on whether the probability is above or below a predetermined threshold. The generation of questions or hypotheses using templates will be described later.

[0059] Furthermore, the generation unit 203 may generate questions and hypotheses without using the generation model 212. For example, the generation unit 203 can generate questions and hypotheses using either or both of pre-created rules and templates. For example, by using the template "What is the use of (a predetermined attribute value extracted from the recommendation reason)?", the generation unit 203 can extract the word "golf," which is the value of the "hobby" attribute, from the recommendation reason in the example of Figure 3, and generate the question "What is the use of golf?". Similarly, by using the template "What kind of person is someone whose (predetermined attribute extracted from the recommendation reason) is (the value of that attribute)?", the generation unit 203 can extract the word "7 million," which is the value of the "annual income" attribute, from the recommendation reason in the example of Figure 3, and generate the question "What kind of person is someone with an annual income of 7 million?". The same applies to hypotheses. For example, the generation unit 203 can use the template "(Recommended item) is related to (a predetermined attribute value extracted from the recommendation reason)" to generate the hypothesis "Product A is related to golf" from the recommended item and recommendation reason in the example in Figure 3.

[0060] Rules for generating questions and hypotheses include, for example, word substitution. For instance, one could define a rule to replace the word "golf" with the broader concept of "sports," or to replace the attribute value "annual income" with "high-income earner," "middle-income earner," etc., depending on the range of the value. By applying such rules, it becomes possible to generate more general questions and hypotheses. For example, instead of, or in addition to, the question "What is the use of golf?", it becomes possible to generate a more general question such as "What is the use of sports?".

[0061] Another example of a rule for generating questions and hypotheses is to select a template to use depending on the attributes used for generation. For example, you could establish a rule that the template "(Hobbies) are useful for generating questions about the attribute "Hobbies" and the template "What kind of person is someone with an annual income of (annual income attribute value)?" are used for generating questions about the attribute "Annual income".

[0062] When generating questions or hypotheses using rules or templates, you can use information related to the recommended product as material for the questions or hypotheses. For example, this could include the attributes of the recommended product (e.g., product name, product category, price or price range, target age, etc.), the reasons for the recommendation, and the attributes of the target person or recommender (e.g., age, gender, occupation, income, etc.).

[0063] As described above, the generation unit 203 may generate questions or hypotheses based on the recommendation reasons generated by the recommendation reason generation unit 202. This provides the effect of being able to present information that delves deeper into the recommendation reasons, in addition to the effects achieved by the information processing device 1 according to the exemplary embodiment 1.

[0064] Furthermore, as described above, the recommendation unit 201 may determine recommended items according to the attributes of the target person using a predictive model 211 generated by learning the relationship between a person's attributes and the results of recommending recommended items to a person with those attributes, or the relationship between a person's attributes and the recommended items that should be recommended to a person with those attributes. In this case, the generation unit 203 may generate questions or hypotheses according to the accuracy of the prediction results of the predictive model 211. This provides the effect of being able to present information that takes into account the accuracy of the prediction results, in addition to the effects of the information processing device 1 according to the exemplary embodiment 1.

[0065] Furthermore, as described above, the generation unit 203 may generate questions or hypotheses based on the attributes of the subject. This provides the effect of being able to present information tailored to the subject, in addition to the effects of the information processing device 1 according to the exemplary embodiment 1.

[0066] Furthermore, as described above, the generation unit 203 may generate questions or hypotheses based on the attributes of the recommender who recommends the recommended items to the target person. This provides the effect of being able to present information that is suitable for the recommender, in addition to the effects of the information processing device 1 according to the exemplary embodiment 1.

[0067] (Answer generation and hypothesis testing methods) This section describes the details of the method used by the response unit 204 to generate answers to questions and to test hypotheses. Answers to questions can be generated, for example, by using a corpus outside the information processing device 2. A corpus is a large-scale collection of structured natural language texts. In this case, the response unit 204 detects from among the many question sentences contained in the corpus that are the same as, or similar in content to, the question sentences generated by the generation unit 203, and generates the corresponding answer sentences as answers to the questions generated by the generation unit 203. Answers can also be generated in the same way using a knowledge graph instead of a corpus. A knowledge graph is a graph structure that systematically links various pieces of knowledge.

[0068] On the other hand, hypothesis testing can be performed using premise sentences whose content is known to be correct and a language understanding model. A language understanding model is a model constructed to output an implication score, which is an index value indicating the degree to which a premise sentence entails the hypothesis sentence when a pair of hypothesis sentences and premise sentences is input. Such a language understanding model can be constructed by learning whether or not a premise sentence entails a hypothesis sentence, using pairs of premise sentences and hypothesis sentences with known implication relationships as training data.

[0069] For example, the response unit 204 may input the hypothesis sentences and premise sentences generated by the generation unit 203 into a language understanding model for various premise sentences, and if any of them have an implication score above a threshold, it may determine that the hypothesis in that hypothesis sentence is correct.

[0070] Furthermore, the method for determining the degree of implication is not limited to the method described above, which utilizes a language understanding model constructed using training data. For example, the response unit 204 may calculate the similarity between the premise sentence and the hypothesis sentence, which have been vectorized by the pre-trained language model, and use the calculated similarity as an index value indicating the degree of implication.

[0071] Furthermore, any method can be used to determine the degree of implication, as long as it allows for the definition of the relationship between the hypothesis and the premise. For example, existing methods such as keyword matching or TF-IDF (Inverse Document Frequency) can be used to determine the degree of implication.

[0072] (Regarding the presentation of answers and verification results) The presentation unit 205 may present the answers to the questions generated by the generation unit 203, or the results of hypothesis testing generated by the generation unit 203, as they are, or it may generate presentation data using the answers or testing results and present the generated presentation data. The method of generating the presentation data is not particularly limited, and for example, presentation data may be generated using predetermined rules or templates. For example, if the above-mentioned answers or testing results are words, the presentation unit 205 may embed those words into a template to form sentences, which can then be used as presentation data. For example, by using the template "(Value of attribute "hobby") is useful for (words included in the question answer)", the presentation unit 205 can present the answer "Networking and health promotion" in the example in Figure 3 as the sentence "Golf is useful for networking and health promotion".

[0073] Furthermore, the presentation unit 205 may generate text corresponding to the response or verification result using a text generation model that has been trained, for example, by learning effective explanations and phrases used by salespeople as training data. This makes it possible for even inexperienced salespeople to conduct effective sales activities.

[0074] Furthermore, it is desirable that the presentation unit 205 presents not only the answers or verification results, but also the corresponding questions and hypotheses, the recommended product, and the reasons for the recommendation. In addition, the presentation unit 205 may also present various information about the recommended product (for example, the specifications of the recommended product, images showing the appearance of the recommended product, reviews about the recommended product, etc.) and the attributes of the target person.

[0075] (Accepting questions regarding the answers and verification results provided) The response unit 204 may accept input of questions regarding the answers or verification results presented by the presentation unit 205. The response unit 204 may then generate an answer to the input question, and the presentation unit 205 may also present that answer. In addition to the effects of the information processing device 1 according to exemplary embodiment 1, this provides the effect of interactively resolving the doubts of the person inputting the question (e.g., the recommender or the subject) by utilizing the response unit 204, which is originally intended to generate answers to questions generated by the generation unit 203.

[0076] The question can be entered via the input unit 22. For example, if the input unit 22 is a device that accepts character input such as a keyboard, the question can be entered as character input. Alternatively, the input unit 22 may be a device such as a speaker that accepts voice input, in which case the question can be entered as voice input, and the voice input can be converted into character data by the information processing device 2 or an external device. Of course, the question can also be entered by an external device of the information processing device 2, and the response unit 204 can acquire the input received by that device via the communication unit 24.

[0077] (Information presentation process) The flow of the information presentation method executed by the information processing device 2 will be explained with reference to Figure 6. Figure 6 is a flowchart showing the flow of the information presentation method according to this exemplary embodiment.

[0078] In S21, the recommendation unit 201 acquires attribute data of the target person. This attribute data is used to determine the target person for recommendation and indicates the attributes of the target person. Subsequently, in S22, the recommendation unit 201 determines the target person for recommendation based on the attribute data acquired in S21. As explained in the section on "Method for determining the target person for recommendation and method for generating recommendation reasons," the recommendation unit 201 may also determine the target person for recommendation based on the output values ​​of the prediction model 211. The presentation unit 205 may also present the target person for recommendation when it is determined in S22.

[0079] In S23, the recommendation reason generation unit 202 generates the recommendation reasons for the recommended target determined in S22. The method for generating the recommendation reasons is as described in the section "Method for Determining Recommended Targets and Method for Generating Recommendation Reasons". The presentation unit 205 may also present the recommendation reasons when they are generated in S23.

[0080] In S24, the generation unit 203 generates questions or hypotheses corresponding to the recommendation targets determined in S22. As explained in the section on "Method for generating questions and hypotheses," the generation unit 203 may generate questions or hypotheses using the generation model 212.

[0081] In S25, the response unit 204 generates an answer to the question generated in S24, or the result of verifying the hypothesis generated in S24. The method for generating the answer and the verification result is as described in the section "Answer Generation and Hypothesis Verification Method".

[0082] In S26, the presentation unit 205 presents the answer or verification result generated in S25. Preferably, in addition to the answer or verification result generated in S25, the presentation unit 205 also presents the question or hypothesis corresponding to the answer or verification result. Furthermore, if the recommendation target and recommendation reason have not been presented in S22 and S23, the presentation unit 205 may also present the recommendation target and recommendation reason in S26.

[0083] In S27, the response unit 204 determines whether a question regarding the question presented in S26 or a question about the verification result has been entered. If the result in S27 is YES, the process returns to S25, the response unit 204 generates an answer to the entered question, and in the subsequent S26, the presentation unit 205 presents that answer. On the other hand, if the result in S27 is NO, the process in Figure 6 ends.

[0084] As described above, the information presentation method according to this exemplary embodiment includes generating questions or hypotheses corresponding to the recommendation target determined to be recommended to the target person (S24), and presenting answers to the generated questions or the results of verifying the generated hypotheses (S26). Therefore, this information presentation method has the effect of making it possible to present useful information when recommending the recommendation target.

[0085] [Variation] The entities executing each process described in the exemplary embodiments above are arbitrary and not limited to the examples given. In other words, an information presentation system having the same functions as the information processing device 2 can be constructed using multiple devices that can communicate with each other. For example, an information presentation system having the same functions as the information processing device 2 can be constructed by distributing each block shown in Figure 4 across multiple devices. Furthermore, each process in the flowchart of Figure 6 can be divided and executed by multiple processors.

[0086] [Examples of implementation using software] Some or all of the functions of information processing devices 1 and 2 may be implemented by hardware such as integrated circuits (IC chips), or by software.

[0087] In the latter case, information processing devices 1 and 2 are implemented, for example, by a computer that executes instructions for a program, which is software that implements each function. An example of such a computer (hereinafter referred to as computer C) is shown in Figure 7. Computer C comprises at least one processor C1 and at least one memory C2. Memory C2 stores a program (information presentation program) P that causes computer C to operate as information processing device 1 or 2. In computer C, the processor C1 reads program P from memory C2 and executes it, thereby implementing each function of information processing device 1 or 2.

[0088] For processor C1, for example, a CPU (Central Processing Unit), GPU (Graphic Processing Unit), DSP (Digital Signal Processor), MPU (Micro Processing Unit), FPU (Floating Point Number Processing Unit), PPU (Physics Processing Unit), TPU (Tensor Processing Unit), quantum processor, microcontroller, or a combination thereof can be used. For memory C2, for example, flash memory, HDD (Hard Disk Drive), SSD (Solid State Drive), or a combination thereof can be used.

[0089] Computer C may also be equipped with RAM (Random Access Memory) for loading program P at runtime and for temporarily storing various data. Furthermore, computer C may be equipped with communication interfaces for sending and receiving data with other devices. Additionally, computer C may be equipped with input / output interfaces for connecting input / output devices such as keyboards, mice, displays, and printers.

[0090] Furthermore, program P can be recorded on a non-temporary, tangible recording medium M that is readable by computer C. Such a recording medium M could be, for example, tape, disk, card, semiconductor memory, or programmable logic circuitry. Computer C can acquire program P via such a recording medium M. Program P can also be transmitted via a transmission medium. Such a transmission medium could be, for example, a communication network or broadcast waves. Computer C can also acquire program P via such a transmission medium.

[0091] [Additional Note 1] The present invention is not limited to the embodiments described above, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technical means disclosed in the embodiments described above are also included in the technical scope of the present invention.

[0092] [Additional Note 2] Some or all of the embodiments described above may also be described as follows. However, the present invention is not limited to the embodiments described below.

[0093] (Note 1) An information processing device comprising: a generation means for generating questions or hypotheses corresponding to the recommended subject determined to be recommended to the subject; and a presentation means for presenting answers to the questions generated by the generation means, or the results of verifying the hypotheses generated by the generation means.

[0094] (Note 2) The information processing device according to Appendix 1, comprising a recommendation reason generation means for generating the recommendation reasons for the recommended subject, wherein the generation means generates the question or hypothesis based on the recommendation reasons.

[0095] (Note 3) The information processing device according to Appendix 1 or 2, comprising a recommendation means for determining the recommended items according to the attributes of the target person, using a predictive model generated by learning the relationship between the attributes of a person and the result of recommending recommended items to a person with those attributes, or the relationship between the attributes of a person and the recommended items that should be recommended to a person with those attributes, wherein the generation means generates the question or the hypothesis based on the accuracy of the prediction result of the predictive model.

[0096] (Note 4) The generation means is an information processing device according to any one of the appendices 1 to 3, which generates the question or hypothesis based on the attributes of the subject.

[0097] (Note 5) The generation means is an information processing device according to any one of the appendices 1 to 4, which generates the question or hypothesis based on the attributes of the recommender who recommends the recommended subject to the subject.

[0098] (Note 6) The information processing device according to any one of Appendix 1 to 5, wherein the generation means generates the question or hypothesis using a generative model generated by learning the relationship between the recommendation target and the question or hypothesis corresponding to the recommendation target.

[0099] (Note 7) An information processing device according to any one of the appendices 1 to 6, comprising a response means for generating an answer to the question generated by the generation means, or a verification result of the hypothesis generated by the generation means, wherein the response means generates an answer to the input question when a question regarding the answer or verification result presented by the presentation means is input, and the presentation means presents the answer to the input question generated by the response means.

[0100] (Note 8) An information presentation method comprising: at least one processor generating questions or hypotheses corresponding to recommendation targets determined to be recommended to a target person; and presenting answers to the generated questions or the results of verifying the generated hypotheses.

[0101] (Note 9) An information presentation program that causes a computer to function as a generating means for generating questions or hypotheses corresponding to recommendation targets determined to be recommended to a target person, and as a presenting means for presenting answers to the questions generated by the generating means, or the results of verifying the hypotheses generated by the generating means.

[0102] [Additional Note 3] Some or all of the embodiments described above can also be expressed as follows: An information processing device comprising at least one processor, the processor performing the processes of generating questions or hypotheses corresponding to recommendation targets determined to be recommended to a target person, and presenting answers to the generated questions or the results of verifying the generated hypotheses.

[0103] Furthermore, this information processing device may also be equipped with memory, which may store an information presentation program that causes the processor to perform the process of generating a question or hypothesis and the process of presenting an answer or verification result. This information presentation program may also be recorded on a computer-readable, non-temporary, tangible recording medium. [Explanation of Symbols]

[0104] 1, 2 Information Processing Devices 201 Recommendation department (recommendation means) 202 Recommendation reason generation unit (recommendation reason generation means) 11, 203 Generation unit (generation means) 204 Response unit (response means) 12, 205 Presentation unit (presentation means) 211 Predictive Models 212 Generative Models

Claims

1. A generation means for generating questions or hypotheses corresponding to the recommended subjects that have been determined to be recommended to the target subjects, The system comprises a presentation means for presenting answers to questions generated by the generation means, or the results of verifying hypotheses generated by the generation means, The generation means generates the question or hypothesis using a generative model generated by learning the relationship between the recommendation target and the question or hypothesis corresponding to the recommendation target. Information processing device.

2. The system includes a recommendation reason generation means for generating the recommendation reasons for the aforementioned recommended target, The information processing apparatus according to claim 1, wherein the generation means generates the question or hypothesis based on the recommendation reason.

3. The system includes a recommendation means that determines the recommended items according to the attributes of the target person using a predictive model generated by learning the relationship between a person's attributes and the results of recommending items to a person with those attributes, or the relationship between a person's attributes and the recommended items that should be recommended to a person with those attributes. The information processing apparatus according to claim 1 or 2, wherein the generation means generates the question or hypothesis according to the accuracy of the prediction result of the prediction model.

4. The information processing apparatus according to claim 1 or 2, wherein the generation means generates the question or hypothesis based on the attributes of the subject.

5. The information processing apparatus according to claim 1 or 2, wherein the generation means generates the question or hypothesis based on the attributes of the recommender who recommends the recommended subject to the subject.

6. The system includes a response means that generates an answer to the question generated by the generation means, or a result of verifying the hypothesis generated by the generation means, The response means generates an answer to the input question when a question regarding the answer or verification result presented by the presentation means is input. The information processing apparatus according to claim 1 or 2, wherein the presentation means presents the answer to the input question, which is generated by the response means.

7. At least one processor, To generate questions or hypotheses that correspond to the recommended subjects, which have been determined to be suitable for recommendation to the target individuals, This includes presenting answers to the generated questions or the results of testing the generated hypotheses, In generating the aforementioned questions or hypotheses, a generative model is used, which is generated by learning the relationship between the recommended target and the questions or hypotheses corresponding to that target. Information presentation method.

8. Computers, A means for generating questions or hypotheses corresponding to the recommendation target that has been determined to be recommended to the target person, and It functions as a presentation means that presents answers to questions generated by the generation means, or the results of verifying hypotheses generated by the generation means. The generation means is an information presentation program that generates the question or hypothesis using a generative model generated by learning the relationship between the recommendation target and the question or hypothesis corresponding to the recommendation target.

Images