Information processing method, information processing device, and program

Abstract

To provide an information processing method, an information processing device, and a program that can perform appropriate information recommendation at an introduction destination facility even if data of a user's action history on items at the introduction destination facility cannot be used to evaluate the performance of a model.SOLUTION: The information processing method is performed by one or more processors. A plurality of models trained using one or more of datasets including a user's action history on items collected at each of a plurality of mutually different first facilities are prepared. The information processing method includes acquiring, by the one or more processors, characteristics of a second facility different from the plurality of first facilities and each of the plurality of first facilities, assessing a similarity between the acquired characteristics of the second facility and the characteristics of the first facility for which the datasets used to train the models are collected, and selecting a model suitable for the second facility from among the plurality of models on the basis of the similarity.SELECTED DRAWING: Figure 11

Description

Technical Field

[0001] The present disclosure relates to an information processing method, an information processing apparatus, and a program, and particularly relates to an information recommendation technique for performing robust recommendations against domain shift.

Background Art

[0002] In a system that provides various items to users (consumers), such as an EC (Electronic Commerce) site or a document information management system, it is difficult for a user to select the best item suitable for themselves from many items both in terms of time and cognitive ability. Items in an EC site are products handled by the EC site, and items in a document information management system are document information stored in the system.

[0003] In order to assist users in selecting items, an information recommendation technique that presents selection candidates from many items has been studied. Generally, when introducing a recommendation system into a certain facility or the like, the model of the recommendation system is learned based on data collected at the facility where the system is introduced. However, when the same recommendation system is introduced into a facility different from the facility where the data used for learning was collected, there is a problem that the prediction accuracy of the model decreases. The problem that a machine learning model does not function well in an unknown other facility is called domain shift, and regarding domain generalization, which is research on improving robustness against domain shift, in recent years, research has been actively conducted mainly in the field of image recognition. However, there are still few research cases on domain generalization in information recommendation techniques.

[0004] Non-Patent Literature 1 describes a method for selecting a model to be used for transfer learning, i.e., a pre-trained model for fine-tuning, from among several models trained in several different languages, in interlanguage transfer learning applied to cross-language translation. In Non-Patent Literature 1, the similarity between the target domain and the source domain is estimated based on several features in order to perform interlanguage transfer learning. The features used to estimate similarity include dataset size, word overlap, geographic distance, genetic distance, and phonological distance.

[0005] Non-Patent Document 2 describes a configuration for predicting user item ratings where, when data from a target domain and multiple source domains are available, the predictions based on the source domain data are weighted by the similarity between the source domain and the target domain and then added together. The domain similarity in Non-Patent Document 2 is configured to learn from the data in a way that minimizes the prediction error for the target domain.

[0006] Patent Document 1 describes a configuration in which the similarity between a set of pre-stored models and the patient's feature data is determined, and a model with features similar to the target patient's feature data is searched for and used from among the set of models. [Prior art documents] [Patent Documents]

[0007] [Patent Document 1] Patent No. 6782802 [Non-patent literature]

[0008] [Non-Patent Document 1] Yu-Hsiang Lin, Chian-Yu Chen, Jean Lee, Zirui Li, Yuyan Zhang,Mengzhou Xia, Shruti Rijhwani, Junxian He, Zhisong Zhang, Xuezhe Ma, Antonios Anastasopoulos, Patrick Littell, Graham Neubig “Choosing Transfer Languages ​​for Cross-Lingual Learning”(ACL 2019) [Non-Patent Document 2] “Transfer Learning for Collective Link Prediction in Multiple Heterogenous Domains” (ICML 2010) [Non-Patent Document 3] Ivan Cantador, Ignacio Fenandez-Tobias, Shlomo Bwrkovsky, Paolo Cremonesi, Chapter 27:"Cross-domain Recommender System" (2015 Springer) [Overview of the project] [Problems that the invention aims to solve]

[0009] Non-patent document 1 is a study on translation technology, not on information recommendation technology. In the technology described in Non-patent document 1, the similarity estimation performance deteriorates if language-specific features are not included in the similarity estimation.

[0010] The technology described in Non-Patent Document 2 cannot learn domain similarity without historical data on user behavior or evaluations of the target domain.

[0011] Furthermore, both Non-Patent Document 1 and Non-Patent Document 2 are studies on domain adaptation and are not aimed at generalization to unknown domains.

[0012] The technology described in Patent Document 1 is applicable to patient-specific models, but not to facility-specific (domain-specific) models as envisioned in information recommendation. Furthermore, even if the patient features are similar, the model's predictive performance may not be sufficient due to differences in domains.

[0013] When the training domain and the target domain differ, one way to achieve robust information recommendation against domain shifts is to train multiple models using datasets collected from multiple different facilities in advance. Then, before implementation, evaluate the performance of these multiple models using a dataset containing user behavior history for items collected at the target facility, and select the optimal model from among them.

[0014] However, during the pre-implementation model evaluation, it may be impossible to prepare a dataset from the target facility, or a sufficient amount of data necessary for model evaluation may not be available. In such cases, it becomes impossible to evaluate the performance of pre-prepared candidate models and select a model suitable for the target facility.

[0015] This disclosure is made in light of these circumstances and aims to provide an information processing method, information processing device, and program that enable information recommendation using a model suitable for the facility where the product is implemented, even when user behavior history data for items at the facility cannot be used to evaluate the model's performance. [Means for solving the problem]

[0016] An information processing method relating to a first aspect of this disclosure is an information processing method performed by one or more processors, wherein multiple models are prepared, each trained using one or more datasets containing user behavior histories for items collected at each of a plurality of different first facilities, and the one or more processors acquire the characteristics of a second facility different from the plurality of first facilities and each of the plurality of first facilities, evaluate the similarity between the acquired characteristics of the second facility and the characteristics of the first facilities from which the datasets used to train each model were collected, and select a model suitable for the second facility from among the plurality of models based on the similarity.

[0017] According to this embodiment, each of the multiple models provided is a model trained using one or more datasets from among multiple datasets collected at multiple first facilities. Instead of directly evaluating the performance of each model at the second facility, one or more processors evaluate the similarity between facilities using the characteristics of the first facilities from which the datasets used to train each model were collected, and the characteristics of the second facility. A model trained using a dataset collected at a first facility with similar characteristics to those of the second facility as the primary dataset may exhibit relatively high performance at the second facility as well. According to this embodiment, even if user behavior history data for items at the second facility is unavailable, a model suitable for information recommendation at the second facility can be selected based on the similarity of the facility characteristics.

[0018] The term "facility" includes the concept of a group composed of multiple users, such as a company, hospital, store, government agency, or e-commerce site. Each of the multiple first and second facilities can be in a different domain from the others.

[0019] The information processing method according to the second aspect of this disclosure includes, in the information processing method according to the first aspect, one or more processors extracting statistical information from a dataset of metadata which are explanatory variables used to train the model, and the characteristics may be configured to include statistical information.

[0020] In the information processing method according to the third aspect of the present disclosure, in the information processing method according to the second aspect, the metadata may be configured to include at least one of user attributes and item attributes.

[0021] In the information processing method according to the fourth aspect of the present disclosure, in the information processing method according to any one of the first to third aspects, one or more processors obtain facility-related information other than the metadata included in the dataset used for learning the model, and the characteristics may be configured to include the facility-related information.

[0022] In the information processing method according to the fifth aspect of the present disclosure, in the information processing method according to the fourth aspect, the facility-related information may be configured to be extracted by web crawling.

[0023] In the information processing method according to the sixth aspect of the present disclosure, in the information processing method according to the fourth or fifth aspect, one or more processors may be configured to receive facility-related information via a user interface.

[0024] In the information processing method according to the seventh aspect of the present disclosure, in the information processing method according to any one of the first to sixth aspects, one or more processors obtain an evaluation value of the prediction performance at a first facility where the datasets used for learning each of the plurality of models are collected, and select a model suitable for a second facility from the plurality of models based on the similarity and the evaluation value of the prediction performance.

[0025] In the information processing method according to the eighth aspect of the present disclosure, in the information processing method according to any one of the first to seventh aspects, one or more processors obtain fitness evaluation information indicating an evaluation of the fitness of the model for the second facility separately from the similarity, and select a model suitable for the second facility from the plurality of models based on the similarity and the fitness evaluation information. '

[0026] The information processing method relating to the ninth aspect of this disclosure may also include, in the information processing method relating to the eighth aspect, the conformity assessment information including the results of a questionnaire given to users of the second facility.

[0027] The information processing method relating to the tenth aspect of this disclosure may include, in an information processing method relating to any one aspect of the first to ninth aspects, one or more processors that use the characteristics of a plurality of third facilities whose similarity to the characteristics of the first facility has been evaluated, and evaluate the similarity between the characteristics of the second facility and the characteristics of the first facility based on the similarity between the characteristics of the second facility and the characteristics of the plurality of third facilities.

[0028] The information processing method according to the 11th aspect of this disclosure may include a configuration in which one or more processors store in a storage device the characteristics of a plurality of third facilities and the similarity between the characteristics of the first facility and the characteristics of the plurality of third facilities.

[0029] The information processing method relating to the twelfth aspect of this disclosure is an information processing method relating to any one aspect of the first to eleventh aspects, wherein the model may be a predictive model used in a recommendation system that recommends items to a user.

[0030] The information processing method relating to the 13th aspect of this disclosure may include a configuration in which one or more processors store multiple models in a storage device, in an information processing method relating to any one of the 1st to 12th aspects.

[0031] The information processing method according to the 14th aspect of this disclosure may include, in the information processing method according to the 13th aspect, a configuration in which one or more processors store in a memory device the characteristics of the first facility from which the datasets used to train each model were collected, in association with the model.

[0032] An information processing device according to a 15th aspect of this disclosure is an information processing device comprising one or more processors and one or more storage devices that store instructions to be executed by the one or more processors, wherein multiple models trained using one or more datasets containing user behavior histories for items collected at each of a plurality of different first facilities are stored in the storage device, and one or more processors acquire the characteristics of a second facility different from the plurality of first facilities and each of the plurality of first facilities, evaluate the similarity between the acquired characteristics of the second facility and the characteristics of the first facilities from which the datasets used to train each model were collected, and select a model suitable for the second facility from among the plurality of models based on the similarity.

[0033] In the 15th embodiment of the information processing apparatus, the configuration may include specific embodiments similar to those of the information processing method relating to any one of the second to 14th embodiments described above.

[0034] The program according to the 16th aspect of this disclosure provides a computer with the following functions: a function to store multiple models trained using one or more datasets containing user behavior histories for items collected at each of several different first facilities; a function to acquire the characteristics of a second facility different from the several first facilities and the characteristics of each of the several first facilities; a function to evaluate the similarity between the acquired characteristics of the second facility and the characteristics of the first facilities from which the datasets used to train each model were collected; and a function to select a model suitable for the second facility from among several models based on the similarity.

[0035] In the program of the 16th embodiment, the configuration may include specific embodiments similar to those of the information processing method relating to any one of the second to 14th embodiments described above. [Effects of the Invention]

[0036] According to this disclosure, even if user behavior history data for items at a second facility, which is different from the first facility where the dataset used to train the model was collected, cannot be used to evaluate the model's performance, a model suitable for the second facility can be selected from among multiple models. This makes it possible to perform appropriate information recommendations at the second facility using the selected model. [Brief explanation of the drawing]

[0037] [Figure 1] Figure 1 is a conceptual diagram of a typical recommendation system. [Figure 2] Figure 2 is a conceptual diagram illustrating an example of supervised machine learning, which is widely used in building recommendation systems. [Figure 3] Figure 3 is an explanatory diagram showing a typical implementation flow of a recommendation system. [Figure 4] Figure 4 is an explanatory diagram of the implementation flow of the recommendation system when data from the target facility is not available. [Figure 5] Figure 5 is an explanatory diagram for training a model through domain adaptation. [Figure 6] Figure 6 is an explanatory diagram of the recommendation system implementation flow, which includes a step to evaluate the performance of the trained model. [Figure 7] Figure 7 is an explanatory diagram showing examples of training data and evaluation data used in machine learning. [Figure 8] Figure 8 is a graph that schematically shows the differences in model performance due to differences in datasets. [Figure 9] Figure 9 is an explanatory diagram illustrating an example of a recommendation system implementation flow when the learning domain and the target domain are different. [Figure 10] Figure 10 is an explanatory diagram illustrating the challenges that arise when there is no user behavior history available at the facility where the system is implemented. [Figure 11] Figure 11 is an explanatory diagram showing an overview of the information processing method according to the first embodiment. [Figure 12]Figure 12 is an explanatory diagram showing an example of processing performed by the information processing device according to the embodiment. [Figure 13] Figure 13 is a block diagram schematically showing an example of the hardware configuration of an information processing device. [Figure 14] Figure 14 is a functional block diagram showing the functional configuration of the information processing device. [Figure 15] Figure 15 is a flowchart illustrating an example of the operation of an information processing device. [Figure 16] Figure 16 is an explanatory diagram illustrating an example of extracting statistical information on user attributes as a characteristic of a facility. [Figure 17] Figure 17 is an explanatory diagram illustrating an example of extracting statistical information on item attributes as characteristics of a facility. [Figure 18] Figure 18 is an explanatory diagram illustrating an example of extracting information about a facility using web crawling. [Figure 19] Figure 19 is an explanatory diagram showing an overview of the information processing method according to the second embodiment. [Figure 20] Figure 20 is an explanatory diagram that schematically shows the characteristics of each facility in a vector space that represents the characteristics of the facilities. [Figure 21] Figure 21 is an example of a directed acyclic graph (DAG) that represents the interdependencies between variables in a joint probability distribution P(X,Y). [Figure 22] Figure 22 shows a specific example of a probability representation of a conditional probability distribution P(Y|X). [Figure 23] Figure 23 is an explanatory diagram showing the relationship between the equation that expresses the conditional probability of a user's action toward an item (Y=1) for a given combination of user behavior characteristics and item characteristics, and the DAG that represents the interdependencies between variables in the joint probability distribution P(X,Y). [Figure 24] Figure 24 is an explanatory diagram showing the relationship between user behavior characteristics defined by the combination of user attribute 1 and user attribute 2, item behavior characteristics defined by the combination of item attribute 1 and item attribute 2, and a DAG that represents the dependencies between variables. [Modes for carrying out the invention]

[0038] Preferred embodiments of the present invention will be described below with reference to the attached drawings.

[0039] An Overview of Information Recommendation Technology First, we will provide an overview of information recommendation technology, including its basic concepts and challenges, with specific examples. Information recommendation technology is a technology used to recommend (suggest) items to users.

[0040] Figure 1 is a conceptual diagram of a typical recommendation system 10. The recommendation system 10 accepts user information and context information as input and outputs information about items to recommend to the user according to the context. Context refers to various "situations," such as the day of the week, time of day, or weather. Items can be various objects, such as books, videos, or restaurants.

[0041] Recommendation system 10 typically recommends multiple items simultaneously. Figure 1 shows an example where recommendation system 10 recommends three items, IT1, IT2, and IT3. Generally, a recommendation is considered successful if the user gives a positive response to the recommended items IT1, IT2, and IT3. Positive responses include, for example, purchase, viewing, or visit. Such recommendation technologies are widely used, for example, on e-commerce sites and gourmet websites that introduce restaurants.

[0042] The recommendation system 10 is built using machine learning techniques. Figure 2 is a conceptual diagram showing an example of supervised machine learning widely used in building the recommendation system 10. Generally, positive and negative examples are prepared based on the user's past behavior history, and combinations of users and contexts are input into the prediction model 12. The prediction model 12 is then trained to minimize the prediction error. For example, viewed items are used as positive examples, and unviewed items are used as negative examples. Machine learning is performed until the prediction error converges, and the target prediction performance is achieved.

[0043] Using the trained prediction model 12, the system recommends items that are predicted to have a high probability of being viewed for a given combination of user and context. For example, when a combination of user A and context β is input to the trained prediction model 12, the prediction model 12 infers that user A is highly likely to view a document like item IT3 under the conditions of context β, and recommends items similar to item IT3 to user A. Depending on the configuration of the recommendation system 10, it is also common to recommend items to users without considering the context.

[0044] [Examples of data used in the development of recommendation systems] User behavior history is essentially equivalent to "ground truth data" in machine learning. Strictly speaking, it can be understood as a task setting where the next (unknown) action is inferred from past behavior history, but it is more common to learn latent features based on past behavior history.

[0045] User activity history could include, for example, book purchase history, video viewing history, or restaurant visit history.

[0046] Furthermore, the main features include user attributes and item attributes. User attributes can include a variety of elements such as gender, age, occupation, family structure, and residential area. Item attributes can include a variety of elements such as book genre, price, video genre, length, restaurant genre, and location.

[0047] [Model construction and operation] Figure 3 is an explanatory diagram illustrating a typical implementation flow of a recommendation system. Here, we show a typical flow for implementing a recommendation system in a certain facility. The implementation of a recommendation system involves first building a model 14 that performs the desired recommendation task (Step 1), and then implementing and operating the built model 14 (Step 2). "Building" the model 14, in the case of a machine learning model, includes training the model 14 using training data to create a predictive model (recommendation model) that satisfies practical recommendation performance levels. "Operating" the model 14 means, for example, obtaining an output of a list of recommended items from the trained model 14 in response to an input of a combination of user and context.

[0048] Building Model 14 requires training data. As shown in Figure 3, the recommendation system Model 14 is generally trained based on data collected at the facility where it is implemented. By training with data collected from the facility, Model 14 learns the behavior of users at the facility and can predict recommended items with high accuracy for those users.

[0049] However, due to various circumstances, data from the facility where the system is to be implemented may not be available. For example, in the case of document information recommendation systems in a company's internal system or a hospital's in-house system, the company developing the recommendation model often does not have access to data from the facility where it is to be implemented. If data from the facility where the system is to be implemented is unavailable, the model must instead be trained on data collected from different facilities.

[0050] Figure 4 is an explanatory diagram of the recommendation system implementation flow when data from the target facility is unavailable. Model 14, which was trained using data collected from a facility different from the target facility, Guidance When implemented at a client's facility, the prediction accuracy of Model 14 suffers from a decrease due to differences in user behavior between facilities.

[0051] The problem of machine learning models not functioning well in unknown facilities different from the facilities in which they were trained can be broadly understood as a technical challenge to improve robustness to domain shifts, where the source domain in which Model 14 was trained differs from the target domain to which Model 14 is applied. Related to domain generalization is domain adaptation, a method of training using data from both the source and target domains. The purpose of using data from a different domain, even though data from the target domain exists, is to compensate for the insufficient amount of data in the target domain for training.

[0052] Figure 5 is an explanatory diagram for training Model 14 through domain adaptation. Although the amount of data collected at the target domain, the facility where the model is implemented, is relatively smaller in proportion to the amount of data collected at different facilities, by training with both sets of data, Model 14 becomes capable of predicting the behavior of users at the target facility with a certain degree of accuracy.

[0053] [Domain description] The differences in "facilities" mentioned above are a type of domain difference. Non-patent literature 3 (Ivan Cantador et al, Chapter 27: "Cross-domain Recommender System"), which is a document on research into domain adaptation in information recommendation, classifies domain differences into the following four categories.

[0054] [1] Item attribute level: For example, comedy movies and horror movies are in different domains.

[0055] [2] Item type level: For example, movies and television series are in different domains.

[0056] [3] Item level: For example, movies and books are in different domains.

[0057] [4] System level: For example, movies shown in movie theaters and movies shown on television are in different domains.

[0058] The differences in "facilities" shown in Figure 5, etc., correspond to the [4] system-level domain among the four classifications mentioned above.

[0059] When a domain is formally defined, it is specified by the joint probability distribution P(X,Y) of the dependent variable Y and the independent variable X, and d1 and d2 are in different domains when Pd1(X,Y) ≠ Pd2(X,Y).

[0060] The joint probability distribution P(X,Y) can be expressed as the product of the distribution of the explanatory variable P(X) and the conditional probability distribution P(Y|X), or as the product of the distribution of the dependent variable P(Y) and the conditional probability distribution P(Y|X).

[0061] P(X,Y)=P(Y|X)P(X)=P(X|Y)P(Y) Therefore, if one or more of P(X), P(Y), P(Y|X), and P(X|Y) change, it becomes a different domain.

[0062] [Typical patterns of domain shift] [Covariate Shift] When the distributions P(X) of explanatory variables differ, this is called a covariate shift. For example, if the distribution of user attributes differs between datasets, or more specifically, if the male-female ratio differs, this falls under the category of covariate shift.

[0063] [Prior probability shift] When the distribution P(Y) of the dependent variable differs, it is called a prior probability shift. For example, when the average viewing rate or average purchase rate differs between datasets, this falls under the category of prior probability shift.

[0064] [Concept Shift] A concept shift occurs when the conditional probability distributions P(Y|X) and P(X|Y) are different. For example, if the probability that a company's research and development department reads a data analysis document is P(Y|X), and this probability differs across datasets, this is considered a concept shift.

[0065] Studies on domain adaptation or domain generalization can be broadly categorized into those that assume one of the above patterns as the primary cause, and those that consider how to deal with changes in P(X,Y) without specifically considering which pattern is the primary cause. In the former case, covariate shifts are particularly common.

[0066] [Reasons why domain shifts have an impact] Predictive / classification models that perform prediction or classification tasks make inferences based on the relationship between the explanatory variable X and the target variable Y. Therefore, if P(Y|X) changes, the prediction / classification performance will naturally decrease. Furthermore, when machine learning predictive / classification models are developed, the model minimizes the prediction / classification error within the training data. For example, when the frequency of the explanatory variable X=X_1 is greater than the frequency of X=X_2, i.e., P(X=X_1)>P(X=X_2), there is more data for X=X_1 than for X=X_2. Therefore, error reduction for X=X_1 takes precedence over error reduction for X=X_2. Consequently, prediction / classification performance also decreases when P(X) changes between facilities.

[0067] Domain shifts can be a problem for models of various tasks, not just information recommendation. For example, a model that predicts employee turnover risk may become problematic when a predictive model trained on data from one company is used in another company.

[0068] Furthermore, with regard to models that predict antibody production by cells, domain shift can become a problem when a model trained using data from one antibody is applied to a different antibody. Similarly, with regard to models that classify customer feedback (VOC), such as a model that categorizes VOC into "product features," "support," and "other," domain shift can become a problem when a classification model trained using data from one product is applied to a different product.

[0069] [Regarding pre-implementation evaluation of the model] Before introducing the trained model 14 to actual facilities, it is common practice to evaluate the model 14's performance. This performance evaluation is necessary for determining whether or not to introduce the model, and for research and development of the model or training method.

[0070] Figure 6 is an explanatory diagram of the recommendation system implementation flow, which includes a step to evaluate the performance of the trained model 14. In Figure 6, a step to evaluate the performance of model 14 is added as "Step 1.5" between Step 1 (the step to train model 14) and Step 2 (the step to operate model 14) as explained in Figure 5. The other configurations are the same as in Figure 5. As shown in Figure 6, in a typical recommendation system implementation flow, the data collected at the facility where the system is to be implemented is often divided into training data and evaluation data. After confirming the predictive performance of model 14 using the evaluation data, operation of model 14 is started.

[0071] However, when constructing a domain generalization model 14, the training data and evaluation data must be from different domains. Furthermore, in domain generalization, it is preferable to use training data from multiple domains, and it is even preferable to have more domains available for training.

[0072] [Regarding generalization] Figure 7 is an explanatory diagram illustrating examples of training data and evaluation data used in machine learning. A dataset obtained from the joint probability distribution Pd1(X,Y) of a certain domain d1 can be divided into training data and evaluation data. Evaluation data from the same domain as the training data is called the "first evaluation data" and is denoted as "evaluation data 1" in Figure 7. In addition, a dataset obtained from the joint probability distribution Pd2(X,Y) of a different domain d2 is prepared and used as evaluation data. Evaluation data from a different domain than the training data is called the "second evaluation data" and is denoted as "evaluation data 2" in Figure 7.

[0073] Model 14 is trained using training data from domain d1, and the performance of the trained model 14 is evaluated using the first evaluation data from domain d1 and the second evaluation data from domain d2.

[0074] Figure 8 is a schematic graph illustrating the difference in model performance due to differences in datasets. If we denote the performance of Model 14 in the training data as Performance A, the performance of Model 14 in the first evaluation data as Performance B, and the performance of Model 14 in the second evaluation data as Performance C, then typically, as shown in Figure 8, the relationship is Performance A > Performance B > Performance C.

[0075] The high generalization performance of Model 14 generally refers to a high performance B, or a small difference between performance A and B. In other words, it aims for high predictive performance even on untrained data without overfitting to the training data.

[0076] In the context of domain generalization as used herein, this refers to a high performance C, or a small difference between performance B and performance C. In other words, the goal is to achieve consistently high performance even in domains different from those used for training.

[0077] Even if behavioral history data from the target facility cannot be used during training, if data can be prepared before implementation, it is conceivable that multiple models could be trained using data collected at a different facility than the target facility, and the performance of these multiple models could be evaluated using data collected at the target facility before implementation. Based on the evaluation results, the optimal model could be selected from among the multiple models and applied to the target facility. An example of this is shown in Figure 9.

[0078] Figure 9 is an explanatory diagram illustrating an example of a recommendation system implementation flow when the training domain and the implementation domain are different. As shown in Figure 9, multiple models can be trained using data collected at a facility different from the implementation facility. Here, we show an example in which models M1, M2, and M3 are trained using datasets DS1, DS2, and DS3, each collected at a different facility. For example, model M1 is trained using dataset DS1, model M2 is trained using dataset DS2, and model M3 is trained using dataset DS3. Note that the datasets used to train each of models M1, M2, and M3 may be a combination of multiple datasets collected at different facilities. For example, model M1 may be trained using a dataset that is a mixture of datasets DS1 and DS2.

[0079] After training multiple models M1, M2, and M3, the performance of each model M1, M2, and M3 is evaluated using data Dtg collected at the facility where the system will be implemented. In Figure 9, the symbols "A," "B," and "C" shown below each model M1, M2, and M3 represent the evaluation results for each model. An A rating indicates good predictive performance that meets the implementation criteria. A B rating indicates performance that is worse than an A rating. A C rating indicates performance that is even worse than a B rating and is unsuitable for implementation.

[0080] For example, as shown in Figure 9, if the evaluation result for Model M1 is "A", the evaluation result for Model M2 is "B", and the evaluation result for Model M3 is "C", then Model M1 will be selected as the optimal model for the facility to implement the system, and the recommendation system 10 using Model M1 will be implemented.

[0081] [Description of the task] In this embodiment, we assume situations where data on users' behavioral history with items at the target facility cannot be prepared during either model training or pre-implementation evaluation, or where data exists but the amount is insufficient to perform model evaluation.

[0082] Figure 10 is an explanatory diagram illustrating the challenges when user behavior history is not available at the facility where the system is implemented. The data used in information recommendation can be broadly divided into user behavior history and metadata such as user attributes and item attributes. User behavior history is the dependent variable, while metadata such as user attributes and item attributes are independent variables.

[0083] Without data on user behavior history at the facility where the system is implemented, it is impossible to evaluate the performance of the model. In other words, since the model used in the information recommendation system predicts the target variable based on explanatory variables, the prediction accuracy of the model cannot be evaluated without ground truth data for the target variable. If the performance of the model cannot be evaluated, it becomes difficult to select the model that is suitable for the facility from among multiple models. This problem also exists when user behavior history at the facility exists, but the amount of data is small and insufficient to perform a proper model evaluation.

[0084] In this embodiment, even when there is no behavioral history data at the facility where the system is implemented, or when the amount of data is insufficient to evaluate the model, a means is provided to select a high-performance model from among multiple models for the facility where the system is implemented. In the following description, the facility where the recommendation system 10 is implemented will be referred to as the "implementation facility," and the facility that collects data for training the candidate models will be referred to as the "training facility." The implementation facility corresponds to the target domain, and the training facility corresponds to the training domain.

[0085] [Outline of the information processing method according to the first embodiment] Figures 11 and 12 are explanatory diagrams illustrating an overview of the information processing method according to the first embodiment. Here, we describe an example in which Model M1 and Model M2 are prepared in advance as candidate models, but in reality, many more models may be prepared.

[0086] Candidate model M1 is a predictive model trained using data collected at learning facility FA1, which is different from the implementation facility FAt. Similarly, another candidate model M2 is a predictive model trained using data collected at learning facility FA2, which is different from both the implementation facility FAt and learning facility FA1.

[0087] User behavior history at the implemented facility FAt will not be available. "Not available" includes concepts such as when behavior history does not exist, when the data is inaccessible even if it exists, or when the amount of data is insufficient for model evaluation. On the other hand, for the implemented facility FAt shown in Figure 11, a dataset Dmt containing metadata such as user attributes and / or item attributes will be assumed to exist.

[0088] In this case, the information processing device 100 according to this embodiment performs processing according to the following procedure (steps 1 to 3).

[0089] [Step 1] In Step 1, the information processing device 100 extracts information that represents the characteristics of each of the learning facilities FA1, FA2, and the implementation facility FAt. The information that represents the characteristics of learning facilities FA1 and FA2 may be statistical values ​​or distributions extracted from the metadata (explanatory variable) datasets Dm1 and Dm2 used for learning through statistical processing.

[0090] Facility characteristics information, such as statistical information extracted from the metadata of explanatory variables, is called "metadata-derived facility characteristics information." When the explanatory variable is a continuous value, the statistical information as metadata-derived facility characteristics information may be, for example, the mean, or a statistical value such as the standard deviation, or a combination thereof. When the explanatory variable is a discrete value, the statistical information as metadata-derived facility characteristics information may be, for example, the mode, or a probability distribution, or a combination thereof.

[0091] Furthermore, information describing the characteristics of learning facilities FA1 and FA2 may be external information separate from the dataset used for learning. This external information may include, for example, information collected from the internet through web crawling, or statistical values ​​and distributions extracted from such collected information. It may also be information entered via a user interface based on publicly available materials.

[0092] In other words, the methods by which the information processing device 100 acquires information about each facility from sources other than the datasets collected at each facility may include automatic collection and / or extraction by the algorithm of the information processing device 100 or another system using web crawling, etc., or by the operator researching and / or inputting publicly available materials, etc.

[0093] External information other than the dataset, such as this, is facility characteristic information that cannot be extracted from the metadata included in the dataset. Facility characteristic information that cannot be extracted from metadata is called "facility-related information outside of metadata."

[0094] The facility characteristic information acquired by the information processing device 100 for each facility may include both facility characteristic information derived from metadata and facility-related information outside of metadata, or it may include only one of these types of information. In the implementation facility FAt shown in Figure 11, it is assumed that a dataset Dmt of explanatory variables (metadata) such as user attributes and item attributes is provided.

[0095] The information processing device 100 acquires facility characteristic information for learning facility FA1, specifically facility characteristic information ST1 derived from metadata and facility-related information EI1 outside of metadata. The information processing device 100 also acquires facility characteristic information for learning facility FA2, specifically facility characteristic information ST2 derived from metadata and facility-related information EI2 outside of metadata.

[0096] Similarly, the information processing device 100 acquires facility characteristic information STt derived from metadata, such as statistical values ​​and distributions extracted from the metadata dataset Dmt collected from the facility FAt, and facility-related information EIt outside of metadata through web crawling or the like.

[0097] [Step 2] In step 2, the information processing device 100 evaluates the similarity between each learning facility FA1, FA2 and the deployment facility FAt based on the facility characteristic information of each facility that has been acquired. For example, the facility characteristic information of each facility is represented as a multidimensional vector, and the similarity is evaluated by the Euclidean distance between the vectors in the vector space.

[0098] [Step 3] In step 3, the information processing device 100 selects a model trained using data collected from a learning facility with a high similarity to the target facility FAt, based on the similarity obtained in step 2. For example, if the evaluation in step 2 shows that the similarity between learning facility FA1 and the target facility FAt is low, and the similarity between learning facility FA2 and the target facility FAt is high, the information processing device 100 selects model M2 from the candidate models M1 and M2 as the model suitable for the target facility FAt.

[0099] Furthermore, if the information processing device 100 uses only facility characteristic information derived from metadata as the facility characteristic information for each facility, it is unnecessary to acquire the facility-related information EI1, EI2, and EIt, which are outside the metadata, as shown in Figure 11. Also, if the information processing device 100 uses only facility-related information outside the metadata as the facility characteristic information for each facility, it is unnecessary to acquire the facility characteristic information ST1, ST2, and STt, which are derived from metadata, as shown in Figure 11.

[0100] [Relationship between datasets and models at each learning facility] As shown in Figures 11 and 12, the relationship between the datasets collected at each learning facility and the models is based on the fundamental idea of ​​training one model from one dataset, using only the dataset from that single learning facility. In this case, the single dataset used to train the model becomes the primary dataset, and the primary learning facility is the source of this dataset.

[0101] However, it is also possible to train a model using a dataset that combines two or more datasets collected from multiple different learning facilities. For example, if a model is trained using a dataset containing 10,000 behavioral history records collected at learning facility 1 and 100 behavioral history records collected at learning facility 2, the majority of the dataset used to train this model will be from learning facility 1, with the proportion of data from learning facility 2 being relatively small. In such a case, the dataset collected at learning facility 1 is considered the primary dataset, and learning facility 1 can be recognized as the primary learning facility. A model trained under such conditions will have high predictive performance at learning facility 1, which is the primary learning facility. Therefore, for models trained using multiple datasets collected from multiple learning facilities, the decision to adopt the model may be made based on the similarity between the characteristics of the facility where it is implemented and the characteristics of the primary learning facility.

[0102] Furthermore, if there are multiple primary learning facilities for a single model, the adoption or rejection of the model may be determined based on representative values ​​such as the average, maximum, or minimum similarity between the characteristics of the facility where the model is implemented and the characteristics of each primary learning facility. For example, if a model is trained using a dataset containing a set of 5,000 behavioral history records collected at learning facility 1 and a set of 5,000 behavioral history records collected at learning facility 2, the proportion of data from learning facility 1 and learning facility 2 in the total dataset used to train the model are equivalent, making it difficult to identify only one primary learning facility. In such cases, learning facility 1 and learning facility 2 may each be treated as primary learning facilities, and the similarity of facility characteristics for multiple combinations of learning facilities may be evaluated by methods such as calculating the average, maximum, or minimum similarity between the characteristics of the facility where the model is implemented and the characteristics of learning facility 1, and between the facility where the model is implemented and learning facility 2. The adoption or rejection of the model may then be determined based on the evaluation results.

[0103] When a model is trained using a dataset that combines two or more datasets collected from multiple learning facilities, the dataset of the learning facility with the largest proportion of data from that facility to the total amount of data used for training may be designated as the "primary dataset." On the other hand, datasets of learning facilities whose proportion of data from that facility to the total amount of data used for training is below a certain threshold may be excluded from being designated as the "primary dataset." The threshold value may be set appropriately within the scope of the technical purpose of determining whether the contribution of a dataset to training can be considered relatively very small, for example, it may be 10% or 5%. Datasets of learning facilities whose proportion of data from that facility to the total amount of data used for training is equal to or greater than the threshold value may be designated as the "primary dataset."

[0104] Furthermore, when training a model using a dataset that combines two or more datasets collected from multiple learning facilities, the similarity of the facility characteristics for multiple combinations of learning facilities may be evaluated by methods such as calculating a weighted average of the similarity between the characteristics of the target facility and the characteristics of each learning facility, using the proportion of the data from each learning facility to the total amount of data used for training as a weight, and then deciding whether or not to adopt the model based on the evaluation results.

[0105] Overview of Information Processing Equipment Figure 13 is a schematic block diagram showing an example of the hardware configuration of the information processing device 100 according to the embodiment. The information processing device 100 can be realized using computer hardware and software. The physical form of the information processing device 100 is not particularly limited and may be a server computer, a workstation, a personal computer, or a tablet terminal. Here, an example of realizing the processing functions of the information processing device 100 using one computer is described, but the processing functions of the information processing device 100 may also be realized by a computer system composed of multiple computers.

[0106] The information processing device 100 includes a processor 102, a computer-readable medium 104 which is a non-temporary tangible object, a communication interface 106, an input / output interface 108, and a bus 110.

[0107] The processor 102 includes a CPU (Central Processing Unit). The processor 102 may also include a GPU (Graphics Processing Unit). The processor 102 is connected to a computer-readable medium 104, a communication interface 106, and an input / output interface 108 via a bus 110. The processor 102 reads various programs and data stored in the computer-readable medium 104 and executes various processes. The term "program" includes the concept of a program module and includes instructions similar to a program.

[0108] The computer-readable medium 104 is a storage device that includes, for example, a main memory 112 and an auxiliary storage device, a storage 114. The storage 114 is configured using, for example, a hard disk drive (HDD), a solid state drive (SSD), an optical disk, a magneto-optical disk, or semiconductor memory, or an appropriate combination thereof. Various programs and data are stored in the storage 114.

[0109] Memory 112 is used as a workspace for the processor 102 and serves as a temporary storage unit for programs and various data read from storage 114. When a program stored in storage 114 is loaded into memory 112, and the processor 102 executes the program's instructions, the processor 102 functions as a means of performing various processes defined by the program.

[0110] Memory 112 stores various programs and data, such as a facility characteristics acquisition program 130, a similarity evaluation program 132, and a model selection program 134, which are executed by the processor 102.

[0111] The facility characteristics acquisition program 130 is a program that executes a process to acquire information indicating the characteristics of learning facilities and implementation facilities. The facility characteristics acquisition program 130 may acquire information indicating the characteristics of learning facilities by, for example, statistically processing data included in a dataset collected at a learning facility. The facility characteristics acquisition program 130 may also accept input of information indicating the characteristics of facilities through, for example, a user interface, or it may include a web crawling program that automatically collects publicly available information indicating the characteristics of facilities from the internet.

[0112] The similarity evaluation program 132 is a program that performs a process to evaluate the similarity of facility characteristics between the implementation facility and each learning facility based on the facility characteristic information of each facility. The model selection program 134 is a program that performs a process to select a model suitable for the implementation facility from among several candidate models based on the similarity evaluation results.

[0113] Memory 112 includes a facility information storage unit 136 and a candidate model storage unit 138. The facility information storage unit 136 is a storage area that stores facility information, including facility characteristic information for each facility acquired by the facility characteristic acquisition program 130. The facility information storage unit 136 may also include a storage area that stores metadata collected at the installation facility.

[0114] The candidate model storage unit 138 is a memory area that stores multiple trained models, each trained using the respective datasets of multiple learning facilities. The candidate model storage unit 138 may also include a memory area that stores the datasets used to train each model. Furthermore, the candidate model storage unit 138 may also include a memory area that stores facility characteristic information of each learning facility in association with (linked to) the models.

[0115] The communication interface 106 performs communication processing with external devices via wired or wireless connections and exchanges information with external devices. The information processing device 100 is connected to a communication line (not shown) via the communication interface 106. The communication line may be a local area network, a wide area network, or a combination of both. The communication interface 106 can act as a data acquisition unit that accepts input of various data, such as datasets.

[0116] The information processing device 100 may include an input device 152 and a display device 154. The input device 152 and the display device 154 are connected to the bus 110 via an input / output interface 108. The input device 152 may be, for example, a keyboard, mouse, multi-touch panel, or other pointing device, or an audio input device, or an appropriate combination thereof. The display device 154 may be, for example, a liquid crystal display, an organic electro-luminescence (OEL) display, or a projector, or an appropriate combination thereof. The input device 152 and the display device 154 may be configured as an integrated unit, such as a touch panel, or the information processing device 100, the input device 152, and the display device 154 may be configured as an integrated unit, such as a touch panel tablet terminal.

[0117] Figure 14 is a functional block diagram showing the functional configuration of the information processing device 100. The information processing device 100 includes a data acquisition unit 220, a data storage unit 222, a facility characteristics acquisition unit 230, a similarity evaluation unit 240, and a model selection unit 244. The data acquisition unit 220 acquires various data, such as metadata about the facility FAt where the device is installed. The data acquisition unit 220 may also include a communication interface 106.

[0118] The data acquired via the data acquisition unit 220 is stored in the data storage unit 222. The data storage unit 222 includes the destination metadata storage unit 224 and the candidate model storage unit 138. The destination metadata storage unit 224 stores a dataset Dmt of metadata such as user attributes and / or item attributes of the destination facility FAt.

[0119] The candidate model storage unit 138 stores multiple candidate models M1, M2...Mn. The candidate model storage unit 138 may also store the datasets DS1, DS2...DSn used to train each model, associated with the model. Here, it is assumed that datasets DS1 and DS2...DSn were collected from different training facilities, and model Mk (k=1,2...n) is a model trained using dataset DSk collected at training facility k. Note that some training facilities may have agreements to discard the datasets used to train the model after training, and therefore, there may be cases where the datasets used to train the model are not stored.

[0120] The facility characteristics acquisition unit 230 acquires information indicating the characteristics of each of the multiple learning facilities k and the implementation facility FAt. The facility characteristics acquisition unit 230 includes a statistical information extraction unit 232 and a metadata-exclusive facility information extraction unit 234. The statistical information extraction unit 232 performs statistical processing on the metadata included in the dataset DSk of each learning facility k and extracts statistical information such as statistical values ​​and / or distributions.

[0121] The non-metadata facility information extraction unit 234 performs web crawling from the internet and extracts non-metadata facility-related information concerning the target facility. Alternatively, the non-metadata facility information extraction unit 234 may accept information input from an operator via a user interface and obtain non-metadata facility-related information about the target facility.

[0122] The similarity evaluation unit 240 evaluates the similarity between each learning facility k and the implementation facility FAt based on the respective facility characteristic information of each learning facility k and the implementation facility FAt.

[0123] The model selection unit 244 selects a model suitable for the FAt facility from among several models based on the similarity evaluated by the similarity evaluation unit 240.

[0124] [Example of a flowchart] Figure 15 is a flowchart illustrating an example of the operation of the information processing device 100. It is assumed that multiple pre-trained models Mk (k=1,2...n) are available. When the flowchart in Figure 15 is started, in step S111, the processor 102 acquires the characteristics of the learning facility k from which the dataset DSk used to train each of the prepared models Mk was collected.

[0125] In step S112, the processor 102 acquires the characteristics of the FAt of the installation site. Note that the order of processing in steps S111 and S112 may be reversed.

[0126] In step S113, the processor 102 evaluates the degree of similarity between the target facility FAt and each learning facility k based on the characteristics of each facility acquired in steps S111 and S112.

[0127] In step S114, the processor 102 selects a model trained using a dataset collected from facilities with high similarity to the target facility FAt. The processor 102 may extract the model trained using the dataset from the facility with the highest similarity as the optimal model, or, if there are multiple facilities whose similarity is above a threshold, it may extract two or more models trained using the datasets of these facilities that meet the acceptable similarity criteria as models applicable to the target facility. If multiple models applicable to the target facility are extracted, the processor 102 may prioritize them in order of the similarity of the facility characteristics and present the models with the highest similarity, or it may present the models as applicable candidates along with the similarity evaluation results. Information on one or more models selected by the processor 102 is output to the display device 154 or the like as a result of the model selection process. After step S114, the processor 102 terminates the flowchart in Figure 14.

[0128] [Specific application examples] This section describes an example of a recommendation system for retail stores. The data used for training includes behavioral history (purchase history) data, user attributes (age), and item attributes (price) for each of the three stores: Store 1, Store 2, and Store 3. Stores 1 through 3 are each training facilities and are examples of "First Facility" in this disclosure. The goal is to develop a recommendation system for a newly opening store 4, but behavioral history data for Store 4 does not yet exist. On the other hand, the product lineup to be sold at Store 4 has been decided, so a dataset of item attributes exists. In addition, a dataset of user attributes for Store 4 exists because store members are being recruited in advance of the opening. Store 4 is the facility where the system will be implemented and is an example of "Second Facility" in this disclosure.

[0129] Under the above conditions, the information processing device 100 performs the following processing steps 1 to 4.

[0130] [Processing Step 1] Processor 102 extracts the characteristics of each of the stores from store 1 to store 4 from the user attribute and item attribute datasets. For example, the average age of users extracted from the user attribute of each of the stores from store 1 to store 4 is 35, 45, 50, and 40 years old, respectively, from store 1 (see Figure 16). Also, for example, the average price of items extracted from the item attribute of each of the stores from store 1 to store 4 is 500 yen, 300 yen, 600 yen, and 400 yen, respectively, from store 1 (see Figure 17). The store characteristics extracted from the dataset may be statistical values ​​or distributions extracted from metadata such as explanatory variables.

[0131] [Processing Step 2] Next, processor 102 extracts the characteristics of each of the stores from 1 to 4, not from the datasets for each store, but from external information separate from the datasets. Processor 102 obtains facility-related information outside of metadata, such as the floor area of ​​each store and the average household income of the municipality where each store is located (see Figure 18). The floor area of ​​each store from 1 to 4 is, for example, 1000m² in order from store 1. 2 , 1500m 2 , 500m 2 , 2000m 2 Furthermore, the average household incomes of the municipalities where each of the stores from Store 1 to Store 4 is located are, for example, 6 million yen, 4 million yen, 7 million yen, and 5 million yen, respectively. Store characteristics extracted from external information separate from the dataset may include various data about the characteristics of the stores themselves that are not included in the dataset. The floor area of ​​each store and the average household income of the municipality where each store is located are examples of "facility-related information other than metadata" in this disclosure.

[0132] [Processing Step 3] From the multiple types of numerical values ​​representing the characteristics of each store obtained by processing steps 1 and 2, the characteristics of each store are represented by a multidimensional vector. In the example above, processor 102 represents the characteristics of each store as a four-dimensional vector of the average user age, average item price, store floor area, and average household income of the city or town where the store is located. Specifically, the characteristic vectors for stores 1 through 4 are (35,500,1000,600), (45,300,1500,400), (50,600,500,700), and (40,400,2000,500).

[0133] [Processing Step 4] Next, processor 102 calculates the similarity of the characteristics of each store. In order to evaluate the similarity of characteristic vectors in the vector space representing the characteristics of the stores, processor 102 calculates the mean and standard deviation for each dimension in order to make the range of values ​​for each dimension equal, and standardizes the values ​​by subtracting the mean from the value of each dimension and dividing by the standard deviation.

[0134] Then, the processor 102 uses the standardized facility characteristic vectors for each store to calculate the Euclidean distance between each vector of store 4 and the other stores 1-3. The Euclidean distance between vectors is one example of an index (evaluation value) used to assess the similarity of facility characteristics.

[0135] For example, the Euclidean distance between the vectors of store 4 and store 1 is 2.05, the Euclidean distance between the vectors of store 4 and store 2 is 1.55, and the Euclidean distance between the vectors of store 4 and store 3 is 3.55. As a result, it can be seen that store 2 has the highest similarity.

[0136] [Processing Step 5] Based on the similarity evaluation results from processing step 4, processor 102 selects the model trained using the dataset of store 2, which has the highest similarity, as the model suitable for store 4.

[0137] Based on the similarities in store characteristics, it is expected that the user behavioral characteristics regarding items are closest to those of store 4 among stores 1 through 3. Therefore, by introducing a model trained using the store 2 dataset to store 4, it is possible to achieve high-performance information recommendation even in store 4, where there is no yet any user behavioral history.

[0138] [Example of extracting user attribute statistics] Figure 16 is an explanatory diagram illustrating an example of extracting statistical information on user attributes as a characteristic of a facility. Here, a retail store such as a supermarket is shown as a specific example of a facility. As previously mentioned, Stores 1 to 3 are learning facilities, and Store 4 is the implementation facility. The same applies to Figures 17 and 18.

[0139] Figure 16 shows an example of data for "age," one of the user attributes for users of each store (facility) from Store 1 to Store 4.

[0140] Processor 102 calculates the average age of users at each store from a dataset of user attributes for each store, as shown in Figure 16. The average age is an example of statistical information and is one of the pieces of information that indicates the characteristics of each store.

[0141] For example, the average age of users at Store 1, calculated from the user attribute data for Store 1, is 35 years old. Similarly, the average age of users at Store 2, calculated from the user attribute data for Store 2, is 45 years old, the average age of users at Store 3 is 50 years old, and the average age of users at Store 4 is 40 years old.

[0142] Processor 102 may further calculate the standard deviation for each store. Alternatively, processor 102 may calculate a histogram of user ages or an age density distribution, in addition to or instead of the average age.

[0143] [Example of extracting statistical information on item attributes] Figure 17 is an explanatory diagram illustrating an example of extracting statistical information on item attributes as a characteristic of a facility. Figure 17 shows an example of data for "price," one of the item attributes for each of the four stores (facilities) from Store 1 to Store 4.

[0144] As shown in Figure 17, processor 102 calculates the average price of items (in this case, goods) at each store from data on the prices of items handled at each store. The average price is an example of statistical information and is one of the pieces of information that indicates the characteristics of each store.

[0145] For example, the average price of items in Store 1, calculated from the item attribute data of Store 1, is 500 yen. Similarly, the average price of items in Store 2, calculated from the item attribute data of Store 2, is 300 yen, the average price of items in Store 3 is 600 yen, and the average price of items in Store 4 is 400 yen.

[0146] Processor 102 may further calculate the standard deviation for each store. Alternatively, processor 102 may calculate, in addition to or instead of the average price, a histogram of item prices or a price density distribution.

[0147] [Example of extracting facility-related information outside of metadata] Figure 18 is an explanatory diagram illustrating an example of extracting facility information through web crawling. Figure 18 shows an example of extracting facility-related information outside of metadata for each store (facility) from Store 1 to Store 4, namely the floor area of ​​each store and the average household income of the municipality where the store is located. The information processing device 100 crawls information on the internet and obtains information on the floor area of ​​each store 1 to 4 and information on the average household income of the municipality where the store is located. The store's address, size, and the average household income of nearby residents can be characteristics of the store that relate to user behavior at that store.

[0148] Furthermore, the information processing device 100 is not limited to performing the crawling; another information processing device (not shown) may perform web crawling, and the information processing device 100 may acquire the information extracted through the crawling.

[0149] Figure 18 shows an example of a retail store, but the content of the information to be extracted may differ depending on the type of facility being targeted. For example, if the facility is a medical facility such as a hospital, facility characteristic information outside the dataset may include the type of hospital, the size of the hospital, or the types of medical departments it offers. The type of hospital may be categorized by its founding body, such as national hospitals, public hospitals, university hospitals, and general hospitals. Furthermore, hospitals may be categorized by function, such as specialized function hospitals, regional medical support hospitals, and others. The size of the hospital may be categorized by the number of beds, for example.

[0150] [Second Embodiment] Figure 19 is an explanatory diagram showing an overview of the information processing method according to the second embodiment. In Figure 19, elements common to Figure 12 are denoted by the same reference numerals, and redundant explanations are omitted. The preconditions described in Figure 12 are the same in Figure 19. In the second embodiment, step 0 is added before step 1 in Figure 12, and steps 3 and 4 of Figure 19 are included instead of step 3 in Figure 12.

[0151] In the second embodiment, the information processing device 100 processes according to the following procedure (steps 0 to 4).

[0152] [Step 0] In Step 0, the information processing device 100 or other machine learning device evaluates the performance of each model after it has been trained. The machine learning device may be a different computer system from the information processing device 100. The evaluation data used to evaluate the predictive performance of each model may be data collected at the same facility as the dataset used for training. The predictive performance of a model is quantified using an indicator (evaluation value), such as prediction accuracy. The information processing device 100 stores the evaluation value of the predictive performance of each model in association with the model. For example, suppose the evaluation value indicating the predictive performance of model M1 is 0.5, and the evaluation value indicating the predictive performance of model M2 is 0.2.

[0153] [Steps 1 and 2] The processes in Step 1 and Step 2 are the same as in Figure 12.

[0154] [Step 3] In step 3, the information processing device 100 calculates a composite score based on the model's predictive performance and the similarity between facilities. Here, an example is shown where the composite score is calculated by multiplying the predictive performance evaluation value and the similarity, but the mean may be used instead of the product.

[0155] Assuming that the similarity between learning facility FA1 and the implementation facility FAt is 0.6, and the similarity between learning facility FA2 and the implementation facility FAt is 0.8, the composite score of model M1's predictive performance and the similarity of learning facility FA1 is calculated to be 0.3, and the composite score of model M2's predictive performance and the similarity of learning facility FA2 is calculated to be 0.16.

[0156] [Step 4] In step 4, the information processing device 100 selects the model with the higher composite score based on the composite score obtained in step 3. In the example in Figure 19, the information processing device 100 selects model M1, which has the higher composite score among models M1 and M2, as the model suitable for the FAt facility where the system will be installed.

[0157] Thus, the configuration may involve selecting a model not only based on the similarity between facilities, but also by taking into account the predictive performance of each model at each learning facility.

[0158] [Variation] In the second embodiment, an example was described in which a composite score that takes into account the predictive performance of each model is used. However, instead of the predictive performance of the models, or in combination with it, a composite score that takes into account some suitability evaluation value that evaluates the suitability of the models at the FAt facility where the models are implemented may be used. The suitability evaluation value may be, for example, the results of a survey of users at the FAt facility where the models are implemented. A suitability evaluation value based on the results of a survey or the like is an example of "suitability evaluation information" in this disclosure.

[0159] [Use the dataset collected at the learning facility after model training.] inability Examples of how to deal with this situation: Due to contractual provisions regarding the destruction of confidential information, learning facilities are required to destroy data sets collected and characteristic data extracted from those datasets after training, and it is conceivable that they may not be able to retain this data.

[0160] In such cases, it becomes impossible to use metadata or statistical values ​​from metadata included in datasets collected at learning facilities when evaluating the similarity between facilities. An example of how to deal with such a situation is explained using Figure 20.

[0161] Figure 20 is an explanatory diagram schematically showing the characteristics of each facility in a vector space representing the characteristics of the facilities. The facility characteristics LD shown by the dashed lines in Figure 20 are data representing the characteristics of learning facilities that become unavailable, such as being discarded after learning. Learning facilities for which facility characteristics LD are used are called "non-retaining learning facilities". In this case, the information processing device 100 may use multiple facility characteristics Dum1, Dum2, and Dum3 whose similarity to the facility characteristics LD of the non-retaining learning facilities has already been evaluated, and evaluate the similarity between the facility characteristics TG of the target facility and each of the multiple facility characteristics Dum1, Dum2, and Dum3, to evaluate the similarity between the facility characteristics LD of the non-retaining learning facilities and the facility characteristics TG of the target facility. Each of the multiple facility characteristics Dum1, Dum2, and Dum3 may be dummy data.

[0162] The information processing device 100 or other information processing device may generate these multiple facility characteristics Dum1, Dum2, and Dum3 based on the facility characteristics LD of the non-retaining learning facility. The information processing device 100 can store the data of the multiple facility characteristics Dum1, Dum2, and Dum3 in association with their similarity to the facility characteristics LD, instead of the facility characteristics LD of the non-retaining learning facility. This makes it possible to evaluate the similarity to the facility characteristics TG of the target facility even without the facility characteristics LD of the non-retaining learning facility. Each of the multiple facility characteristics Dum1, Dum2, and Dum3 is an example of the "characteristics of the third facility" in this disclosure.

[0163] [Explanation of learning methods] Next, we will explain the model training method. Here, we will use matrix factorization, which is frequently used in information recommendation, as an example. In the following explanation, we will show an example where the information processing device 100 performs the training process, but the device that performs the training process may be a computer system other than the information processing device 100.

[0164] Given a dataset containing the behavioral histories of multiple users for multiple items in a learning facility, processor 102 first learns the dependencies between variables based on this data. More specifically, processor 102 represents users and items as vectors, and uses a model in which the sum of their dot products becomes the behavioral probability, updating the model parameters to minimize the error in behavioral prediction.

[0165] A user's vector representation can be expressed, for example, as the sum of the vector representations of each of the user's attributes. The vector representation of an item is similar. A model that has learned the dependencies between variables corresponds to a representation of the joint probability distribution P(X,Y) between the target variable Y and each explanatory variable X in a given dataset of behavioral history.

[0166] Figure 21 is an example of a directed acyclic graph (DAG) that represents the interdependencies between variables in a joint probability distribution P(X,Y). In Figure 21, an example is shown where four variables, user attribute 1, user attribute 2, item attribute 1, and item attribute 2, are used as explanatory variables X. The relationship between each of these explanatory variables X and the user's behavior towards an item, which is the dependent variable Y, can be represented by a graph like the one in Figure 21.

[0167] During training, for example, a vector representation of the joint probability distribution P(X,Y) is obtained based on the dependencies between variables, such as the DAG shown in Figure 21. The graph in Figure 21 shows that the user's behavior towards an item, which is the target variable, depends on the user's behavioral characteristics and the characteristics of the item. It shows that the user's behavioral characteristics depend on user attribute 1 and user attribute 2, and the characteristics of the item depend on item attribute 1 and item attribute 2.

[0168] As shown in Figure 21, the combination of user attribute 1 and user attribute 2 defines the user's behavioral characteristics. Similarly, the combination of item attribute 1 and item attribute 2 defines the item's characteristics. The user's actions toward an item are then determined by the combination of the user's behavioral characteristics and the item's characteristics.

[0169] In general, the relationship P(X,Y) = P(X) × P(Y|X) holds, and when the graph in Figure 21 is applied to this equation, it is expressed as follows. P(X) = P(User Attribute 1, User Attribute 2, Item Attribute 1, Item Attribute 2) P(Y|X) = P(User's action on the item | User attribute 1, User attribute 2, Item attribute 1, Item attribute 2) P(X,Y) = P(User Attribute 1, User Attribute 2, Item Attribute 1, Item Attribute 2) × P(User's actions towards the item | User Attribute 1, User Attribute 2, Item Attribute 1, Item Attribute 2)

[0170] Furthermore, the graph shown in Figure 21 demonstrates that it can be decomposed into the following elements. P(Y|X) = P(User's behavior towards the item | User's behavioral characteristics, Item's characteristics) × P(User's behavioral characteristics | User attribute 1, User attribute 2) × P(Item's behavioral characteristics | Item attribute 1, Item attribute 2)

[0171] [Examples of probability representations for conditional probability distributions P(Y|X)] For example, the probability that a user views an item (Y=1) can be represented by the sigmoid function of the dot product of the user characteristic vector and the item characteristic vector. This method of representation is called matrix factorization. The reason for using the sigmoid function is that its value ranges from 0 to 1, so the function's value can directly correspond to the probability. However, other functions can also be used in the model representation.

[0172] Figure 22 shows a concrete example of the probability representation of P(Y|X). Equation F22A, shown in the upper part of Figure 22, is an example of an equation that expresses the user characteristic vector θu and the item characteristic vector φi as 5-dimensional vectors by matrix decomposition, and the sigmoid function σ(θu·φi) of their inner product (θu·φi) as the conditional probability P(Y=1|user,item).

[0173] `u` is an index value that distinguishes users. `i` is an index value that distinguishes items. Note that the vector's dimension is not limited to 5; it can be set to an appropriate number of dimensions as a hyperparameter of the model.

[0174] The user characteristic vector θu is expressed as the sum of the user's attribute vectors. For example, as shown in equation F22B in the middle of Figure 22, the user characteristic vector θu is expressed as the sum of user attribute 1 vector and user attribute 2 vector. Similarly, the item characteristic vector φi is expressed as the sum of the item's attribute vectors. For example, as shown in equation F22C in the lower part of Figure 22, the item characteristic vector φi is expressed as the sum of item attribute 1 vector and item attribute 2 vector.

[0175] Figure 23 is an explanatory diagram showing the relationship between equation F22A, which represents the conditional probability of a user's action toward an item (Y=1) for a given combination of user behavior characteristics and item characteristics, and a DAG, which represents the interdependencies between variables in the joint probability distribution P(X,Y). As shown in Figure 23, equation F22A represents the conditional probability of the portion enclosed by the dashed frame FR1 in the DAG shown in Figure 23.

[0176] Figure 24 is an explanatory diagram showing the relationship between user behavior characteristics defined by the combination of user attribute 1 and user attribute 2, item characteristics defined by the combination of item attribute 1 and item attribute 2, and a DAG that represents the dependencies between variables. As shown in Figure 24, equation F22B represents the relationship of the portion enclosed by the dashed frame FR2 in the DAG shown in Figure 24. Also, equation F22C represents the relationship of the portion enclosed by the dashed frame FR3 in the DAG shown in Figure 24.

[0177] The values ​​of each vector shown in Figure 23 are determined by learning from the data (training data) included in the user behavior history dataset for a given domain.

[0178] For example, the vector values ​​are updated using stochastic gradient descent (SGD) so that P(Y=1|user,item) is large for user-item pairs that have viewed the item, and small for user-item pairs that have not viewed the item.

[0179] For the joint probability distribution P(X,Y) shown in Figures 23 and 24, the parameters to be learned from the data are as follows. • User characteristic vector: θu • Item characteristic vector: φi • User attribute 1 vector: Vk_u^1 • User attribute vector 2: Vk_u^2 • Item attribute 1 vector: Vk_i^1 • Item attribute 2 vector: Vk_i^2 However, these parameters satisfy the following relationship. ·θu=Vk_u^1+Vk_u^2 ·φi=Vk_i^1+Vk_i^2 k is an index value that distinguishes attributes. For example, if user attribute 1 has 10 types of departments, user attribute 2 has 6 age levels, item attribute 1 has 20 types of products, and item attribute 2 has 5 types of product prices, then the number of attribute types is 10 + 6 + 20 + 5 = 41, so the possible values ​​of k are from 1 to 41. For example, if k=1, it corresponds to the sales department of user attribute 1, and the index value of user attribute 1 for user u is expressed as k_u^1.

[0180] The values ​​of each vector—user attribute 1 vector Vk_u^1, user attribute 2 vector Vk_u^2, item attribute 1 vector Vk_i^1, and item attribute 2 vector Vk_i^2—are obtained through learning from the training data.

[0181] For example, the log loss shown in equation (1) below can be used as the loss function during training.

[0182] L=-{Y ui log σ(θu·φi)+(1-Y ui ) log (1-σ(θu·φi))}(1) If user u views item i, then Y ui = 1, and the larger the predicted probability σ(θu·φi), the smaller the loss L. Conversely, if user u does not view item i, Y ui = 0, and the smaller σ(θu·φi), the smaller the loss L.

[0183] The parameters of the vector representation are learned so that the loss L mentioned above is minimized. For example, when performing optimization using stochastic gradient descent, one record is randomly selected from all the training data (or one UI pair is selected from all UI pairs if context is not required), the partial derivatives (gradients) of each parameter of the loss function are calculated for the selected record, and the parameters are changed in a direction that minimizes the loss L in proportion to the magnitude of the gradient.

[0184] For example, the parameters of the user attribute vector (Vk_u^1) are updated according to equation (2) below.

[0185]

number

[0186] Generally, among a large number of items, there are overwhelmingly more items with Y=0 than items with Y=1. Therefore, when saving behavioral history data as a table as shown in Figure 19, only Y=1 is retained, and pairs of user u and item i that are not included in the behavioral history data are learned as Y=0. In other words, by saving only the positive examples, negative examples can be easily generated as they are not included in the positive examples.

[0187] [Regarding model representation] Matrix factorization is not the only means of representing the joint probability distribution of explanatory variable X and dependent variable Y. For example, logistic regression or Naive Bayes can be applied instead of matrix factorization. Any predictive model can also be used to represent the joint probability distribution by calibrating the output score to be close to the probability P(Y|X). For example, Support Vector Machines (SVMs), Gradient Boosting Decision Trees (GDBTs), and neural network models of any architecture can also be used.

[0188] [Regarding programs that operate computers] It is possible to record a program that enables a computer to implement some or all of the processing functions of the information processing device 100 on a computer-readable medium, such as an optical disk, magnetic disk, or semiconductor memory, which is a tangible, non-temporary information storage medium, and to provide the program through this information storage medium.

[0189] Alternatively, instead of providing programs by storing them on tangible, non-temporary computer-readable media, it is also possible to provide program signals as a download service using telecommunication lines such as the Internet.

[0190] Furthermore, some or all of the processing functions in the information processing device 100 may be implemented by cloud computing, and it is also possible to provide them as SaaS (Software as a Service).

[0191] [Regarding the hardware configuration of each processing unit] The hardware structure of the processing unit that performs various processes in the information processing device 100, such as the data acquisition unit 220, the facility characteristics acquisition unit 230, the statistical information extraction unit 232, the metadata-external facility information extraction unit 234, the similarity evaluation unit 240, and the model selection unit 244, is, for example, various processors as shown below.

[0192] Various types of processors include general-purpose processors such as CPUs, GPUs, and FPGAs (Field Programmable Gate Arrays), which are programmed logic devices (PLDs) whose circuit configurations can be changed after manufacturing, and dedicated electrical circuits, which are processors with circuit configurations specifically designed to perform particular tasks, such as ASICs (Application Specific Integrated Circuits).

[0193] A single processing unit may be composed of one of these various processors, or it may be composed of two or more processors of the same or different type. For example, a single processing unit may be composed of multiple FPGAs, or a combination of a CPU and an FPGA, or a combination of a CPU and a GPU. Alternatively, multiple processing units may be composed of a single processor. Examples of composing multiple processing units with a single processor include, firstly, a configuration where one or more CPUs and software are combined to form a single processor, and this processor functions as multiple processing units, as is typical of computers such as clients and servers. Secondly, a configuration where a processor is used that realizes the functions of the entire system, including multiple processing units, on a single IC (Integrated Circuit) chip, as is typical of System-on-a-Chip (SoC) systems. Thus, various processing units are configured, in terms of hardware structure, using one or more of the above-mentioned various processors.

[0194] Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit composed of circuit elements such as semiconductor devices.

[0195] [Advantages of the Embodiment] According to the embodiments described above, even if user behavior history data for items at a different facility from the one from which the dataset used to train the model was collected cannot be used to evaluate the model's performance, a model suitable for the facility can be selected from among multiple models based on the similarity of the facility's characteristics.

[0196] According to each embodiment, when the domain of the facility where the data used for model training is collected (training domain) and the domain of the facility where the model is implemented (implementation domain) are different, it becomes possible to provide a recommendation list that is robust to domain shifts.

[0197] [Other application examples] In the embodiments described above, user purchasing behavior in a retail store was used as an example. However, the scope of application of this disclosure is not limited to this example. The technology of this disclosure can be applied to models that predict user behavior for various items, regardless of the application, such as browsing documents in a company, browsing medical images and various documents in a hospital or other medical facility, or watching videos and other content on a content provision site.

[0198] 〔others〕 This disclosure is not limited to the embodiments described above, and various modifications are possible without departing from the spirit of the technical idea of ​​this disclosure. [Explanation of Symbols]

[0199] 10 Recommendation System 12 Predictive Models 14 Models 100 Information Processing Devices 102 processors 104 Computer-readable media 106 Communication Interface 108 Input / Output Interfaces 110 Bus 112 memory 114 storage 130 Facility Characteristics Acquisition Program 132 Similarity Evaluation Programs 134 Model Selection Program 136 Facility Information Storage Department 138 Candidate Model Storage Section 152 Input device 154 Display device 220 Data Acquisition Unit 222 Data Storage Section 224 Implemented metadata storage unit 230 Facility Characteristics Acquisition Department 232 Statistical information extraction section 234 Metadata External Facility Information Extraction Unit 240 Similarity Evaluation Unit 244 Model Selection Section DS1 dataset DS2 dataset DS3 dataset DTG Data Dm1 dataset Dm2 dataset DMT dataset Dum1 Facility Characteristics Dum2 Facility Characteristics Dum3 Facility Characteristics LD facility characteristics TG Facility Characteristics EI1 Facility-related information EI2 Facility-Related Information EIt Facility-Related Information ST1 Facility Characteristics Information ST2 Facility Characteristics Information STt Facility Characteristics Information TG Facility Characteristics F22A type F22B formula F22C formula FR1 frame FR2 frame FR3 frame IT1 Item IT2 Item IT3 Item M1 Model M2 Model Mn Model S111~S114 Steps of processing performed by the information processing device

Claims

1. An information processing method performed by one or more processors, Multiple models are prepared, each trained using one or more datasets containing user behavior history, user attributes, and item attributes collected at each of several different primary facilities. Each of the aforementioned models is The one or more of the aforementioned processors To acquire the characteristics of a second facility different from the aforementioned plurality of first facilities, and of each of the aforementioned plurality of first facilities, The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Selecting a model suitable for the second facility from among the multiple models based on the similarity, Includes, The one or more of the aforementioned processors Statistical processing is performed on the dataset of metadata, which are explanatory variables used in training the aforementioned model, and the dataset of metadata collected from the second facility, and statistical information including at least one of the mean, standard deviation, mode, and probability distribution is extracted. The aforementioned characteristics include the aforementioned statistical information, Information processing methods.

2. The metadata includes at least one of user attributes and item attributes. The information processing method according to claim 1.

3. The one or more of the aforementioned processors This includes obtaining facility-related information other than the aforementioned metadata, The aforementioned characteristics include the facility-related information, The information processing method according to claim 1.

4. An information processing method performed by one or more processors, Multiple models are prepared, each trained using one or more datasets containing user behavior history, user attributes, and item attributes collected at each of several different primary facilities. The one or more of the aforementioned processors To acquire the characteristics of a second facility different from the aforementioned plurality of first facilities, and of each of the aforementioned plurality of first facilities, The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Selecting a model suitable for the second facility from among the multiple models based on the similarity, Includes, The one or more of the aforementioned processors A characteristic of the aforementioned facility is that it acquires external information separate from the aforementioned dataset, The aforementioned external information includes facility-related information that cannot be extracted from the metadata included in the dataset and the metadata collected from the second facility. Information processing methods.

5. The aforementioned facility-related information is extracted by web crawling. The information processing method according to claim 3 or 4.

6. The one or more of the aforementioned processors The facility-related information is received via the user interface. The information processing method according to claim 3 or 4.

7. The one or more of the aforementioned processors The evaluation value of the predictive performance at the first facility where the datasets used for training each of the aforementioned multiple models were collected is obtained. Based on the similarity and the evaluation value of the prediction performance, a model suitable for the second facility is selected from among the multiple models. The information processing method according to any one of claims 1 to 4.

8. The one or more of the aforementioned processors In addition to the similarity mentioned above, suitability assessment information indicating the suitability of the model to the second facility is obtained. Based on the similarity and the suitability assessment information, a model suitable for the second facility is selected from among the multiple models. The information processing method according to any one of claims 1 to 4.

9. An information processing method performed by one or more processors, Multiple models are prepared, each trained using one or more datasets containing user behavior history for items collected at each of several different primary facilities. The one or more of the aforementioned processors To acquire the characteristics of a second facility different from the aforementioned plurality of first facilities, and of each of the aforementioned plurality of first facilities, The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Selecting a model suitable for the second facility from among the multiple models based on the similarity, Includes, The one or more of the aforementioned processors In addition to the similarity mentioned above, suitability assessment information indicating the suitability of the model to the second facility is obtained. Based on the similarity and the suitability assessment information, a model suitable for the second facility is selected from among the multiple models. The conformity assessment information includes the results of a questionnaire given to users of the second facility. Information processing methods.

10. An information processing method performed by one or more processors, Multiple models are prepared, each trained using one or more datasets containing user behavior history for items collected at each of several different primary facilities. The one or more of the aforementioned processors To acquire the characteristics of a second facility different from the aforementioned plurality of first facilities, and of each of the aforementioned plurality of first facilities, The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Selecting a model suitable for the second facility from among the multiple models based on the similarity, Includes, The one or more of the aforementioned processors Using the characteristics of multiple third facilities whose similarity to the characteristics of the first facility has already been evaluated, This includes evaluating the similarity between the characteristics of the second facility and the characteristics of the first facility based on the similarity between the characteristics of the second facility and the characteristics of the plurality of third facilities, Information processing methods.

11. The one or more of the aforementioned processors The characteristics of the aforementioned multiple third facilities, This includes storing the degree of similarity between the characteristics of the first facility and the characteristics of the plurality of third facilities in a memory device. The information processing method according to claim 10.

12. The aforementioned model is a predictive model used in a recommendation system that recommends items to users. The information processing method according to any one of claims 1 to 4 and 9 to 11.

13. The one or more of the aforementioned processors This includes storing the aforementioned multiple models in a memory device. The information processing method according to any one of claims 1 to 4 and 9 to 11.

14. The one or more of the aforementioned processors This includes storing in the memory device the characteristics of the first facility from which the dataset used to train each of the aforementioned models was collected, in association with the model. The information processing method according to claim 13.

15. One or more processors, An information processing apparatus comprising one or more storage devices that store instructions to be executed by one or more of the processors, Multiple models trained using one or more datasets containing user behavior history, user attributes, and item attributes collected at each of several different first facilities are stored in the storage device. The one or more of the aforementioned processors The characteristics of a second facility different from the aforementioned plurality of first facilities and each of the aforementioned plurality of first facilities are obtained. The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Based on the similarity, a model suitable for the second facility is selected from among the multiple models. It is an information processing device, The one or more of the aforementioned processors Statistical processing is performed on the dataset of metadata, which are explanatory variables used in training the aforementioned model, and the dataset of metadata collected from the second facility, and statistical information including at least one of the mean, standard deviation, mode, and probability distribution is extracted. The aforementioned characteristics include the aforementioned statistical information, Information processing device.

16. One or more processors, An information processing apparatus comprising one or more storage devices that store instructions to be executed by one or more of the processors, Multiple models trained using one or more datasets containing user behavior history, user attributes, and item attributes collected at each of several different first facilities are stored in the storage device. The one or more of the aforementioned processors The characteristics of a second facility different from the aforementioned plurality of first facilities and each of the aforementioned plurality of first facilities are obtained. The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Based on the similarity, a model suitable for the second facility is selected from among the multiple models. It is an information processing device, The one or more of the aforementioned processors A characteristic of the aforementioned facility is that it acquires external information separate from the aforementioned dataset, The aforementioned external information includes facility-related information that cannot be extracted from the metadata included in the dataset and the metadata collected from the second facility. Information processing methods.

17. One or more processors, An information processing apparatus comprising one or more storage devices that store instructions to be executed by one or more of the processors, Multiple models trained using one or more datasets containing user behavior history for items collected at each of several different first facilities are stored in the storage device. The one or more of the aforementioned processors The characteristics of a second facility different from the aforementioned plurality of first facilities and each of the aforementioned plurality of first facilities are obtained. The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Based on the similarity, a model suitable for the second facility is selected from among the multiple models. It is an information processing device, The one or more of the aforementioned processors In addition to the similarity mentioned above, suitability assessment information indicating the suitability of the model to the second facility is obtained. Based on the similarity and the suitability assessment information, a model suitable for the second facility is selected from among the multiple models. The conformity assessment information includes the results of a questionnaire given to users of the second facility. Information processing device.

18. One or more processors, An information processing apparatus comprising one or more storage devices that store instructions to be executed by one or more of the processors, Multiple models trained using one or more datasets containing user behavior history for items collected at each of several different first facilities are stored in the storage device. The one or more of the aforementioned processors The characteristics of a second facility different from the aforementioned plurality of first facilities and each of the aforementioned plurality of first facilities are obtained. The similarity between the characteristics of the second facility obtained and the characteristics of the first facility from which the dataset used to train each model was collected is evaluated. Based on the similarity, a model suitable for the second facility is selected from among the multiple models. It is an information processing device, The one or more of the aforementioned processors Using the characteristics of multiple third facilities whose similarity to the characteristics of the first facility has already been evaluated, This includes evaluating the similarity between the characteristics of the second facility and the characteristics of the first facility based on the similarity between the characteristics of the second facility and the characteristics of the plurality of third facilities, Information processing device.

19. A program that causes a computer to execute the information processing method described in any one of claims 1, 4, 9, and 10.

Images