Information processing device, information processing method, and program

Abstract

To provide a recommendation technology capable of coping with a cold start problem while suppressing an increase in cost.SOLUTION: An information processing device includes a first acquisition part for acquiring information about a user, a second acquisition part for acquiring information about an item, one or more conversion parts for performing learning by using training data including data generated by a learned model, and respectively converting the information about the user and the information about the item into a vector, and a prediction part for predicting compatibility between the user and the item with reference to the vector converted by the conversion part.SELECTED DRAWING: Figure 1

Description

[Technical Field]

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

[0002] There are known recommendation systems that recommend items (products, measures, etc.) to a user or identify users who match a certain item. For example, Non-Patent Document 1 discloses a technology that recommends new users and new items through randomized learning. [Prior art documents] [Non-patent literature]

[0003] [Non-Patent Document 1] "Recommendation for New Users and New Items via RandomizedTraining and Mixture-of-Experts Transformation", Ziwei Zhu et.al, SIGIR '20, July 25-30, 2020, Virtual Event, China Summary of the Invention [Problem to be solved by the invention]

[0004] Generally, in a recommendation system, there is a problem (also called the cold start problem) of what kind of recommendation should be made to a new user or a new system with which there is no correlation history (e.g., rating history) between them. Non-Patent Document 1 proposes to address the cold start problem by performing learning including auxiliary representations.

[0005] However, the technique of Non-Patent Document 1 requires the preparation of a sufficient amount of training data including auxiliary representations, which increases the cost, which has been a problem.

[0006] The present disclosure has been made in consideration of the above-mentioned problems, and has an object to provide a recommendation technology that can address the cold start problem while suppressing increases in costs. [Means for solving the problem]

[0007] An information processing device according to one aspect of the present disclosure includes a first acquisition means for acquiring information about a user, a second acquisition means for acquiring information about an item, one or more conversion means that are trained using training data including data generated by a trained model and convert each of the information about the user and the information about the item into a vector, and a prediction means that predicts the compatibility between the user and the item by referring to the vector converted by the conversion means.

[0008] An information processing device according to one aspect of the present disclosure includes a first acquisition means for acquiring at least one of information about a user and information about an item, a second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means, and a learning means for training one or more conversion means for converting each of the user and the item into a vector using training data including the data acquired by the second acquisition means.

[0009] An information processing method according to one aspect of the present disclosure includes acquiring information about a user, acquiring information about an item, converting each of the information about the user and the information about the item into a vector using one or more conversion means trained using training data including data generated by a trained model, and predicting compatibility between the user and the item by referring to the converted vector.

[0010] An information processing method according to one aspect of the present disclosure includes acquiring at least one of information about a user and information about an item, acquiring data generated by a trained model by referring to the acquired information, and training one or more conversion means that convert each of the user and the item into a vector using training data including the acquired data.

[0011] A program according to one aspect of the present disclosure is a program that causes a computer to function as an information processing device, and causes the computer to function as a first acquisition means that acquires information about a user, a second acquisition means that acquires information about an item, one or more conversion means that are trained using training data including data generated by a trained model and convert each of the information about the user and the information about the item into a vector, and a prediction means that predicts the compatibility between the user and the item by referring to the vectors converted by the conversion means.

[0012] A program according to one aspect of the present disclosure is a program that causes a computer to function as an information processing device, and causes the computer to function as a first acquisition means that acquires at least one of information about a user and information about an item, a second acquisition means that acquires data generated by a trained model by referring to the information acquired by the first acquisition means, and a learning means that trains one or more conversion means that convert each of the user and the item into a vector using training data including the data acquired by the second acquisition means. [Effects of the Invention]

[0013] According to the present disclosure, it is possible to provide a recommendation technology that can deal with the cold start problem while suppressing increases in costs. [Brief explanation of the drawings]

[0014] [Figure 1] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 2] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 3] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 4] FIG. 1 is a flow diagram showing the flow of an information processing method according to the present disclosure. [Figure 5] 1 is a block diagram illustrating a configuration of an information processing system according to the present disclosure. [Figure 6] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 7] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 8] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 9] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 10] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 11] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 12] FIG. 1 is a diagram for explaining information processing according to the present disclosure. [Figure 13] 1 is a block diagram illustrating a configuration of an information processing device according to the present disclosure. [Figure 14] FIG. 1 is a block diagram illustrating a hardware configuration of an information processing device according to the present disclosure. DETAILED DESCRIPTION OF THE INVENTION

[0015] The following are examples of embodiments of the present invention. However, the present invention is not limited to the exemplary embodiments shown below, and various modifications are possible within the scope of the claims. For example, embodiments obtained by appropriately combining the technologies (part or all of the products or methods) employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, embodiments obtained by appropriately omitting some of the technologies employed in the exemplary embodiments shown below may also be included in the scope of the present invention. Furthermore, the effects mentioned in the exemplary embodiments shown below are examples of effects expected in the exemplary embodiments, and do not define the scope of the present invention. In other words, embodiments that do not exhibit the effects mentioned in the exemplary embodiments shown below may also be included in the scope of the present invention.

[0016] [First embodiment] A first exemplary embodiment, which is an example of an embodiment of the present invention, will be described in detail with reference to the drawings. This exemplary embodiment is a basic form for each of the exemplary embodiments described below. Note that the scope of application of each technique employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technique employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise. Furthermore, each technique shown in the drawings referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical obstacles arise.

[0017] (Configuration of information processing device 1) The configuration of an information processing device 1 according to this exemplary embodiment will be described with reference to Fig. 1. Fig. 1 is a block diagram showing the configuration of the information processing device 1. As shown in Fig. 1, the information processing device 1 includes a first acquisition unit 11, a second acquisition unit 12, a conversion unit 13, and a prediction unit 14.

[0018] (First acquisition unit 11) The first acquisition unit 11 acquires information about a user. Here, the information about the user may include, for example, a user ID and user attribute information associated with the user ID. Furthermore, the user attribute information may include, for example, the gender and age of the user, and text that describes the user. For example, the user attribute information may include: The user's self-introduction or sales description, for example, "This person likes ____." A description of the user's interests, such as "This person's interests are golf, wine, and raising children." -Descriptions about the products purchased by the user, the advertisements viewed, the stores used, etc. (descriptions about the user's behavioral history), for example, "The items purchased by this person are chocolate confectionery, coffee, and mulled wine," or "The products purchased by this person are characterized as luxury, luxury goods, and European." may be included.

[0019] The format of the user attribute information is not particularly limited, and may be written in a format that can be applied to a predetermined format, or the format may be changed as appropriate using attributes included in the attribute information. For example, UserID item 001 Chocolate confectionery 002 Coffee 003 Mulled wine and data structured as follows: item attribute Chocolate sweets, sweet, high-quality Coffee luxury goods, luxury Mulled wine, a luxury item in Europe The above-mentioned user attribute information may be constructed by associating the above-mentioned data with each other through a relational database.

[0020] However, the above example does not limit the present exemplary embodiment. For example, the user attribute information may include an image representing the user (an image of the user himself / herself, an image of an item the user likes, etc.) instead of or in addition to the text, etc.

[0021] As mentioned above, the user attribute information may include a correlation history between the user and one or more items. Whether or not the user has purchased (used) the item (purchase history, usage history) - What rating the user gave to the item (rating history) It may also include information about:

[0022] (Second acquisition unit 12) The second acquisition unit 12 acquires information about an item. Here, the "item" includes, as an example, at least one of a product and a campaign. The "campaign" also includes the date and time of the campaign, the name, the service or product provided, etc. The information about the item may also include, as an example, an item ID and item attribute information associated with the item ID. Here, the item attribute information includes, as an example, text expressing the content of the item. For example, the item attribute information may include: A description that explains the item (measure), for example, "This is a campaign to introduce XX, which is popular in XX." A description of the item (product), such as "Wine and cheese imported directly from Germany." However, the above example does not limit the present exemplary embodiment. For example, the item attribute information may include an image representing the item (such as an image of the product or an image of the campaign venue) instead of or in addition to the text, etc.

[0023] The item attribute information may also include a correlation history between the item and one or more users. · Whether or not the item has been purchased (used) by the user (purchase history, usage history) - What rating the item received from the user (rating history) It may also include information about:

[0024] (Conversion unit 13) The conversion unit 13 Information about the user acquired by the first acquisition unit 11, and Information about the item acquired by the second acquisition unit 12 into a vector. Here, the conversion unit 13 is trained using training data including data generated by the trained model LM. The conversion unit 13 may also be configured by individual conversion units that convert each of the information about the user and the information about the item into a vector. For example, the conversion unit 13 may be configured by a first conversion unit that converts the information about the user into a vector (also referred to as a user vector) and a second conversion unit that converts the information about the item into a vector (also referred to as an item vector).

[0025] Moreover, the conversion unit 13, for example, Generate the user vector by inputting information about the user into an embedding model (embedding function) f trained using training data including data generated by the trained model LM; Generate the item vector by inputting information about the items into an embedding model (embedding function) f' trained using the training data including data generated by the trained model LM. Here, the embedding model f' may be different from or the same as the embedding model f. Furthermore, the conversion unit 13 may be one that has been trained by contrastive learning. Furthermore, the specific example of the trained model LM does not limit this exemplary embodiment, but as an example, it may be a language model trained to be able to generate text, or a generative model trained to be able to generate images.

[0026] (Prediction Section 14) The prediction unit 14 predicts the compatibility between the user and the item by referring to the vector converted by the conversion unit 13. As an example, the prediction unit 14 Deriving the similarity between the user vector and the item vector generated by the conversion unit 13; Predicting the compatibility between the user and the item by referring to the derived similarity. The prediction result by the prediction unit 14 may be visually presented to the user via a display unit (not shown) or provided to another device via a communication unit (not shown), for example.

[0027] (Effects of information processing device 1) As described above, in the information processing device 1, Obtain information about the user and information about the item; converting each of the information about the user and the information about the item into a vector using one or more conversion means (conversion unit 13) trained using training data including data generated by the trained model LM; Predicting the compatibility between the user and the item by referring to the vector converted by the conversion means The following configuration is adopted.

[0028] In this way, in the information processing device 1, Convert each of the information about the user and the information about the item into a vector. Since the above configuration is adopted, even new users and items can be converted into vectors, and a recommendation technology that can deal with the cold start problem is provided. Using training data including data generated by the trained model LM, a vector conversion means (conversion unit 13) is trained; Predicting the compatibility between the user and the item by referring to the vector converted by the conversion means (conversion unit 13). This configuration allows the conversion means (conversion unit 13) to be suitably trained using data generated by the trained model LM, while eliminating the need to use the trained model LM in the inference phase. Therefore, this configuration provides a recommendation technology that suppresses increases in cost.

[0029] (Flow of information processing method S1) Next, the flow of information processing method S1 according to this exemplary embodiment will be described with reference to Fig. 2. Fig. 2 is a flow diagram showing the flow of information processing method S1. As shown in Fig. 2, information processing method S1 includes step (process) S11 of acquiring information about a user, step (process) S12 of acquiring information about an item, step (process) S13 of converting to a vector, and step (process) S14 of predicting compatibility.

[0030] (Step S11) In step S11, the first acquisition unit 11 acquires information about the user. A more specific description of the first acquisition unit 11 has been given above, so a description thereof will be omitted here.

[0031] (Step S12) In step S12, the second acquisition unit 12 acquires information related to the item. A more specific description of the second acquisition unit 12 has been given above, so a description thereof will be omitted here.

[0032] (Step S13) Next, in step S13, the conversion unit 13 Information about the user acquired by the first acquisition unit 11, and Information about the item acquired by the second acquisition unit 12 into a vector. Here, the conversion unit 13 is trained using training data including data generated by the trained model LM. A more detailed explanation of the conversion unit 13 has been given above, so a detailed explanation will be omitted here.

[0033] (Step S14) Next, in step S14, the prediction unit 14 predicts the compatibility between the user and the item by referring to the vector converted by the conversion unit 13. A more specific description of the prediction unit 14 has been given above, and therefore will not be repeated here.

[0034] (Effect of information processing method S1) As described above, in the information processing method S1, Obtain information about the user and information about the item; converting each of the information about the user and the information about the item into a vector using one or more conversion means (conversion unit 13) trained using training data including data generated by the trained model LM; Predicting the compatibility between the user and the item by referring to the vector converted by the conversion means According to the above configuration, the same effects as those of the information processing device 1 are achieved.

[0035] (Configuration of information processing device 2) Next, the configuration of the information processing device 2 according to this exemplary embodiment will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the information processing device 2. As shown in Fig. 3, the information processing device 2 includes a first acquisition unit 21, a second acquisition unit 22, and a learning unit 23.

[0036] (First acquisition unit 21) The first acquisition unit 21 acquires at least one of information about a user and information about an item. Here, the information about the user may include, for example, user attribute information associated with the user. The information about the user may also include the user ID of the user. The user attribute information may also include, for example, at least one of the user's gender, age, and text describing the user. For example, the user attribute information may include: The user's self-introduction or sales description, for example, "This person likes ____." A description of the user's interests, such as "This person's interests are golf, wine, and raising children." -Descriptions about the products purchased by the user, the advertisements viewed, the stores used, etc. (descriptions about the user's behavioral history), for example, "The items purchased by this person are chocolate confectionery, coffee, and mulled wine," or "The products purchased by this person are characterized as luxury, luxury goods, and European." may be included.

[0037] The format of the user attribute information is not particularly limited, and may be written in a format that can be applied to a predetermined format, or the format may be changed as appropriate using attributes included in the attribute information. For example, UserID item 001 Chocolate confectionery 002 Coffee 003 Mulled wine and data structured as follows: item attribute Chocolate sweets, sweet, high-quality Coffee luxury goods, luxury Mulled wine, a luxury item in Europe The above-mentioned user attribute information may be constructed by associating the above-mentioned data with each other through a relational database.

[0038] However, the above example does not limit the present exemplary embodiment. For example, the user attribute information may include an image representing the user (an image of the user himself / herself, an image of an item the user likes, etc.) instead of or in addition to the text, etc.

[0039] As mentioned above, the user attribute information may include a correlation history between the user and one or more items. Whether or not the user has purchased (used) the item (purchase history, usage history) - What rating the user gave to the item (rating history) It may also include information about:

[0040] On the other hand, the information about the item may include, for example, item attribute information associated with the item. The information about the item may also include an item ID of the item. The item attribute information may also include, for example, text expressing the contents of the item. For example, the item attribute information may include: A description that explains the item (measure), for example, "This is a campaign to introduce XX, which is popular in XX." A description of the item (product), such as "Wine and cheese imported directly from Germany." However, the above example does not limit the present exemplary embodiment. For example, the item attribute information may include an image representing the item (such as an image of the product or an image of the campaign venue) instead of or in addition to the text, etc.

[0041] The item attribute information may also include a correlation history between the item and one or more users. · Whether or not the item has been purchased (used) by the user (purchase history, usage history) - What rating the item received from the user (rating history) It may also include information about:

[0042] (Second acquisition unit 22) The second acquisition unit 22 acquires data generated by the trained model LM by referring to the information (at least one of information about the user and information about the item) acquired by the first acquisition unit 21. As an example, the second acquisition unit 22 Generate one or more prompts by referring to the information acquired by the first acquisition unit 21; Inputting the generated one or more prompts into a trained model LM, and acquiring the data output by the trained model as data generated by the trained model LM. Here, the "prompt" refers to, for example, "data converted into text, which is the input format for the trained model LM." However, this term does not limit this exemplary embodiment.

[0043] Note that the specific example of the trained model LM does not limit this exemplary embodiment, but as an example, it may be a language model trained to be able to generate text, or a generative model trained to be able to generate images.

[0044] (Study Section 23) The learning unit 23 trains one or more conversion means that convert each of the user and the item into a vector using training data including the data acquired by the second acquisition means. Here, the conversion means may be configured with individual conversion units that convert each of the information about the user and the information about the item into a vector. For example, similar to the conversion unit 13 included in the information processing device 1, the conversion means may be configured with a first conversion unit that converts the information about the user into a vector (also referred to as a user vector) and a second conversion unit that converts the information about the item into a vector (also referred to as an item vector).

[0045] Moreover, the conversion means may be, for example: Generate the user vector by inputting information about the user into an embedding model (embedding function) f trained using training data including data generated by the trained model LM; Generate the item vector by inputting information about the items into an embedding model (embedding function) f' trained using the training data including data generated by the trained model LM. Here, the embedded model f' may be different from the embedded model f, or may be the same as the embedded model f.

[0046] The learning unit 23 trains the conversion means configured as described above using training data including the data acquired by the second acquisition means. As an example, the learning unit 23 trains the conversion means by contrastive learning. Furthermore, the learning unit 23 trains the conversion means by: A first loss value obtained by referring to data generated by the trained model LM; a second loss value obtained by referring to data other than the data generated by the trained model LM; The conversion means may be trained to reduce these loss values ​​by referring to the above.

[0047] The conversion means (or parameters defining the conversion means) learned by the learning unit 23 is stored in a storage unit (not shown), for example, and is used in the conversion process in the inference phase.

[0048] (Effects of information processing device 2) As described above, in the information processing device 2, Acquire at least one of information about the user and information about the item; - Refer to the acquired information to acquire data generated by the trained model; One or more conversion means for converting each of the users and the items into vectors are trained using training data including the acquired data. The following configuration is adopted.

[0049] In this way, in the information processing device 2, One or more conversion means for converting each of the users and the items into vectors are trained using training data including the acquired data. This configuration allows the system to learn a conversion method that can convert even new users and items into vectors, providing a recommendation technology that can address the cold start problem. Furthermore, while the conversion method can be optimally trained using data generated by the trained model LM, there is no need to use the trained model LM in the inference phase. Therefore, this configuration provides a recommendation technology that suppresses cost increases.

[0050] (Flow of information processing method S2) Next, the flow of the information processing method S2 according to this exemplary embodiment will be described with reference to Fig. 4. Fig. 4 is a flow diagram showing the flow of the information processing method S2. As shown in Fig. 4, the information processing method S2 includes a step (process) S21 of acquiring at least one of information about a user and information about an item, a step (process) S22 of acquiring data generated by a trained model, and a step (process) S23 of performing learning using training data.

[0051] (Step S21) In step S21, the first acquisition unit 21 acquires at least one of information about the user and information about the item. A more specific description of the first acquisition unit 21 has been given above, and therefore will not be repeated here.

[0052] (Step S22) Subsequently, in step S22, the second acquisition unit 22 acquires data generated by the trained model LM by referring to the information (at least one of information about the user and information about the item) acquired by the first acquisition unit 21 in step S21. A more specific description of the second acquisition unit 22 has been given above, and therefore will not be repeated here.

[0053] (Step S23) Next, in step S23, the learning unit 23 causes one or more conversion means that convert each of the users and the items into vectors to learn using training data including the data acquired by the second acquisition means in step S22. A more specific description of the learning unit 23 has been given above, so a description thereof will be omitted here.

[0054] (Effect of information processing method S2) As described above, in the information processing method S2, Acquire at least one of information about the user and information about the item; - Refer to the acquired information to acquire data generated by the trained model; One or more conversion means for converting each of the users and the items into vectors are trained using training data including the acquired data. The above configuration provides the same effects as the information processing device 2.

[0055] Second Embodiment A second exemplary embodiment, which is one example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technology shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs.

[0056] (Configuration of information processing system 1A) The configuration of an information processing system 1A according to this exemplary embodiment will be described with reference to Fig. 5. Fig. 5 is a block diagram showing the configuration of the information processing system 1A. As shown in Fig. 5, the information processing system 1A includes an information processing device 100 and a generating device 50 connected to the information processing device 100 via a network N. Here, the specific configuration of the network N does not limit this exemplary embodiment, but as an example, a wireless LAN (Local Area Network), a wired LAN, a WAN (Wide Area Network), a public line network, a mobile data communication network, or a combination of these networks can be used.

[0057] (Generation device 50) As shown in FIG. 5 , the generation device 50 includes a control unit 51, a storage unit 52, and a communication unit 53. The communication unit 53 communicates with devices external to the generation device 50. As an example, the communication unit 53 communicates with an information processing device 100 included in the information processing system 1A. The communication unit 53 transmits data supplied from the control unit 51 to the information processing device 100, and supplies data received from the information processing device 100 to the control unit 51. Note that the data received by the communication unit 53 from the information processing device 100 may include a prompt generated by the information processing device 100. Furthermore, the data provided by the communication unit 53 to the information processing device 100 may include at least one of text data and image data generated by a language model LM (described later) based on the prompt.

[0058] A language model LM is stored in the storage unit 52. As an example, the storage unit 52 stores a plurality of parameters that define the language model LM. These parameters are, as an example, parameters that have been trained in advance by machine learning (parameters that have undergone an update process by machine learning), but this does not limit the present exemplary embodiment. Furthermore, as an example, the language model LM is a model that has been trained using a large number of texts and images, but this does not limit the present exemplary embodiment.

[0059] The control unit 51 uses the language model LM to obtain an output result from the language model LM. As an example, the control unit 51 inputs a prompt received from the information processing device 100 into the language model LM and obtains text data or image data generated by the language model LM. The control unit 51 also provides the text data or image data to the information processing device 100 via the communication unit 53.

[0060] In the present exemplary embodiment, the generation device 50 is illustrated as a device separate from the information processing device 100, but this does not limit the present exemplary embodiment. The control unit 51 included in the generation device 50 or the function of the language model execution unit in the control unit 51 may be configured to be included in the control unit of the information processing device 100. Similarly, the language model LM2 stored in the storage unit 52 included in the generation device 50 may be stored in the storage unit of the information processing device 100, and the information processing device 100 itself may be configured to be able to execute the language model LM.

[0061] (Configuration of information processing device 100) Next, the configuration of the information processing device 100 according to this exemplary embodiment will be described with reference to Fig. 5. As shown in Fig. 5, the information processing device 100 includes a control unit 10, a storage unit 20, a communication unit 30, and an input / output unit 40.

[0062] (Communication unit 30) The communication unit 30 communicates with devices external to the information processing device 100. As an example, the communication unit 30 communicates with the generation device 50. The communication unit 53 transmits data supplied from the control unit 10 to the generation device 50, and supplies data received from the generation device 50 to the control unit 10. Note that the data transmitted by the communication unit 30 to the generation device 50 may include a prompt generated by the control unit 10. Furthermore, the data received by the communication unit 30 from the generation device 50 may include at least one of text data and image data generated by the generation device 50 based on the prompt.

[0063] (Input / output section 40) The input / output unit 40 is configured to include at least one of input / output devices such as a keyboard, a mouse, a display, a printer, and a touch panel. Alternatively, the input / output unit 40 may be configured to have input / output devices such as a keyboard, a mouse, a display, a printer, and a touch panel connected to it. In this configuration, the input / output unit 40 accepts various types of information input to the information processing device 100 from the connected input devices. Furthermore, the input / output unit 40 outputs various types of information to connected output devices under the control of the control unit 10. An example of the input / output unit 40 is an interface such as a USB (Universal Serial Bus).

[0064] (Storage unit 20) The storage unit 20 stores various data referenced by the control unit 10 and various data generated by the control unit 10. As an example, the storage unit 20 stores: User-related information URI Item related information IRI Vector Information VI Prediction result PRED Prompt PR etc. are stored.

[0065] Here, the user-related information URI is information related to one or more users, and may include, for example, a user ID for each user and user attribute information associated with the user ID. The user attribute information may include, for example, the user's gender, age, and text describing the user. Specific examples of user-related information URIs will be described later, so a detailed description is omitted here.

[0066] On the other hand, an item-related information IRI is information relating to one or more items. Here, the above-mentioned "item" includes, as an example, at least one of a product and a campaign. Furthermore, the "campaign" includes the date and time of the campaign, the name, the service or product provided, etc. Furthermore, the information relating to the item may include, as an example, an item ID and item attribute information associated with the item ID. Here, the item attribute information includes, as an example, text expressing the contents of the item. Specific examples of item-related information IRIs will be described later, so explanation will be omitted here.

[0067] The vector information VI is information about a vector generated by the conversion unit 13, which will be described later. As an example, the vector information VI includes values ​​of each component that defines the vector generated by the conversion unit 13. Specific examples of the vector information VI will be described later, so a description thereof will be omitted here.

[0068] The prediction result PRED is information indicating a prediction result derived by the prediction unit 14, which will be described later. Specific examples of the prediction result PRED will be described later, and therefore will not be described here.

[0069] The prompt PR is a prompt that is generated by the learning unit 23, which will be described later, and is provided to the trained model LM. Specific examples of the prompt PR will be described later, so a description thereof will be omitted here.

[0070] (Control unit 10) 5, the control unit 10 includes an acquisition unit 11, a conversion unit 13, a prediction unit 14, a learning unit 23, and an output unit 15. Here, the acquisition unit 11 also has the same functions as the first acquisition unit 11 and the second acquisition unit 12 included in the information processing device 1 described in exemplary embodiment 1, and the first acquisition unit 21 and the second acquisition unit 22 included in the information processing device 2 described in exemplary embodiment 1, and therefore the acquisition unit 11 may also be referred to as acquisition unit 11 (12, 21, 22).

[0071] (Acquisition part 12) The acquisition unit 12 acquires information about the user and information about the item. Here, the information about the user may include, for example, user attribute information associated with the user, as described in the first exemplary embodiment. The information about the user may also include the user ID of the user. The user attribute information may also include, for example, at least one of the user's gender, age, and text describing the user. For example, the user attribute information may include: The user's self-introduction or sales description, for example, "This person likes ____." A description of the user's interests, such as "This person's interests are golf, wine, and raising children." -Descriptions about the products purchased by the user, the advertisements viewed, the stores used, etc. (descriptions about the user's behavioral history), for example, "The items purchased by this person are chocolate confectionery, coffee, and mulled wine," or "The products purchased by this person are characterized as luxury, luxury goods, and European." may be included.

[0072] The format of the user attribute information is not particularly limited, and may be written in a format that can be applied to a predetermined format, or the format may be changed as appropriate using attributes included in the attribute information. For example, UserID item 001 Chocolate confectionery 002 Coffee 003 Mulled wine and data structured as follows: item attribute Chocolate sweets, sweet, high-quality Coffee luxury goods, luxury Mulled wine, a luxury item in Europe The above-mentioned user attribute information may be constructed by associating the above-mentioned data with each other through a relational database.

[0073] However, the above example does not limit the present exemplary embodiment. For example, the user attribute information may include an image representing the user (an image of the user himself / herself, an image of an item the user likes, etc.) instead of or in addition to the text, etc.

[0074] As mentioned above, the user attribute information may include a correlation history between the user and one or more items. Whether or not the user has purchased (used) the item (purchase history, usage history) - What rating the user gave to the item (rating history) It may also include information about:

[0075] On the other hand, the information about the item may include, for example, item attribute information associated with the item, as described in the first exemplary embodiment. The information about the item may also include an item ID of the item. The item attribute information may also include, for example, text expressing the contents of the item. For example, the item attribute information may include: A description that explains the item (measure), for example, "This is a campaign to introduce XX, which is popular in XX." A description of the item (product), such as "Wine and cheese imported directly from Germany." However, the above example does not limit the present exemplary embodiment. For example, the item attribute information may include an image representing the item (such as an image of the product or an image of the campaign venue) instead of or in addition to the text, etc.

[0076] The item attribute information may also include a correlation history between the item and one or more users. · Whether or not the item has been purchased (used) by the user (purchase history, usage history) - What rating the item received from the user (rating history) It may also include information about:

[0077] (Conversion unit 13) The conversion unit 13 converts into vectors each of the information about the user and the information about the item acquired by the acquisition unit 11. Here, the conversion unit 13 is trained using training data including data generated by the trained model LM described above.

[0078] 6 is a diagram schematically illustrating the processing by the conversion unit 13. As shown in FIG. 6, the conversion unit 13 performs the following processing, for example: Converting information about each of the one or more users included in the user-related information URI into a user vector using an embedding model (embedding function) f; Convert information about each of the one or more items included in the item-related information IRI into an item vector using an embedding model (embedding function) f'. Here, the embedding models f and f' are models learned by a learning unit 23 (described later) using training data including data generated by the learned model LM. As a specific example of the embedding models f and f', a model such as BERT (Bidirectional Encoder Representations from Transformers) can also be used, but this example does not limit the present exemplary embodiment.

[0079] In the example shown in FIG. 6, user-related information (text) for each user identified by user ID=i (i=1 to n) u1, ,u n However, by the embedding model f, each user vector f(u1),···,f(u n ) where each user vector f(u i ) is an example of a d-dimensional vector.

[0080] Similarly, in the example shown in FIG. 6, item-related information (text) for each item identified by item ID=j (j=1 to m) s1, ,s m However, by the embedding model f', each item vector f(s1),...,f(s m ) where each item vector f'(u j ) is an example of a d-dimensional vector.

[0081] The embedding model f' may be different from or the same as the embedding model f. In addition, the embedding model f' may be f'(s) = f(s + "people who are interested in ~") As shown above, a function obtained by generating user text representing a user from text representing an item and adding the user text to the argument of the embedding model f may be used as the embedding model f'.

[0082] Fig. 7 shows a more specific example of processing by the conversion unit 13. In the example shown in Fig. 7, the user-related information URI includes user attribute information about multiple users identified by user IDs = 001 to 999, and each piece of user attribute information includes the user's gender, age, and text describing the user.

[0083] As an example, the conversion unit 13 converts user 001, who has user attribute information of “fashion lover, app lover, seeks delicious food, person with attributes of...”, into a user vector [0.8, -0.5, -0.2, ...] using the embedding model f described above.

[0084] As an example, the conversion unit 13 converts user 002, who has user attribute information of “a person who likes movies, wine, books, etc.”, into a user vector [0.7, -0.1, 0.9, ...] using the embedding model f described above.

[0085] Similarly, in the example shown in Figure 7, the item-related information IRI includes item attribute information for multiple items identified by item IDs = 001 to 099, and each item attribute information includes text that describes the item.

[0086] As an example, the conversion unit 13 converts an item 001 having item attribute information "Thai Food Fair - Offering traditional Thai cuisine and modern cuisine that makes extensive use of spices and seasonings" into an item vector [0.1, 0.5, -0.4, ...] using the embedding model f' described above.

[0087] As an example, the conversion unit 13 converts an item 002 having item attribute information of "German Food Fair - various dishes such as traditional German wine and sausages will be served, accompanied by live German music," into an item vector [0.4, -0.1, 0.7, ...] using the embedding model f' described above.

[0088] The vector converted (generated) by the conversion unit 13 is stored in the storage unit 20 as vector information VI.

[0089] (Prediction Section 14) The prediction unit 14 predicts the compatibility between the user and the item by referring to the vector converted by the conversion unit 13. As an example, the prediction unit 14 Deriving the similarity between the user vector and the item vector generated by the conversion unit 13; Predicting the compatibility between the user and the item by referring to the derived similarity. For example, the prediction unit 14 may calculate the similarity sim(u, s) between the user vector f(u) generated by the conversion unit 13 and the item vector f′(s) as follows: sim(u,s)=<f(u),f’(s)> It can be calculated using the inner product as shown below,

number

[0090] If the similarity calculated as described above is equal to or greater than a predetermined threshold, the prediction unit 14 sets the value of the reaction flag to 1 (determines that the compatibility is good), and if not, sets the value of the reaction flag to 0 (determines that the compatibility is bad). Note that a specific method for setting the threshold will be described later.

[0091] Fig. 8 is a diagram showing a specific example of the prediction result PRED derived by the prediction unit 14 with reference to the similarity. In the example shown in Fig. 8, the prediction result PRED includes information indicating the compatibility between a certain item (item 001) and each user. More specifically, the example shown in Fig. 8 shows an example in which the above-mentioned threshold is 0.65, User 001, whose similarity to item 001 is 0.7, has a reaction flag of 1. User 002, whose similarity to item 001 is 0.4, has a reaction flag of 0. User 999, whose similarity to item 001 is 0.8, has a reaction flag of 1. The prediction result is derived as shown below. The threshold value referred to by the prediction unit 14 may be set in advance, or may be derived (set) by the prediction unit 14 based on training data. The process of deriving the threshold value by the prediction unit 14 will be described later.

[0092] (Output section 15) The output unit 15 presents the prediction result PRED derived by the prediction unit 14 to the user via the input / output unit 40. As an example, the output unit 15 may be configured to visually present the prediction result PRED shown in FIG. 8 to the user via the input / output unit 40.

[0093] The output unit 15 may also be configured to present at least a portion of a prompt PR (for example, prompts PR1 to PR1 described later) generated by the learning unit 23 described later, or at least a portion of the data generated by the trained model LM with reference to the prompt PR, to the user via the input / output unit 40. The output unit 15 may also be configured to accept, via the input / output unit 40, a correction instruction from the user regarding the prompt PR and at least a portion of the data generated by the trained model LM, and to correct the target data based on the correction instruction.

[0094] (Study Section 23) The learning unit 23 trains the conversion unit 13 using training data. Here, the training data includes data generated by a trained model LM that references at least one of information about the user and information about the item acquired by the acquisition unit 11. As an example, the learning unit 23 may train the conversion unit 13 by contrastive learning. For example, the learning unit 23 trains the conversion unit 13 by - To increase the similarity between the item and the positive examples related to that item, -The similarity between the item and the negative examples related to that item is low. Learning may be performed by contrastive learning. As an example, the learning unit 23 may perform learning by contrastive learning using one or more of various algorithms such as SimCSE, Dense Passage Retrieval, and sentence-BERT.

[0095] In the above process, the type of data to be used as a positive example or a negative example does not limit the present exemplary embodiment. For example, Actual text from users who have given the above item a rating of at least a certain value (for example, a rating of 4 or higher on a 1-5 scale), or Data generated by the trained model LM as text from users who gave the above item a rating above a predetermined value The following can be used as the positive examples. The actual text of users who gave the above item a rating below a certain value (e.g., below 3 on a rating scale of 1 to 5), or Data generated by the trained model LM as text from users who gave the item a rating below a predetermined value etc. can be used as the negative examples.

[0096] Furthermore, the learning unit 23 controls the conversion unit 13 as follows: A first loss value obtained by referring to data generated by the trained model LM; a second loss value obtained by referring to data other than the data generated by the trained model LM; The conversion means may be trained to reduce these loss values ​​by referring to the above. In the following description, data generated by the trained model LM may be referred to as a pseudo sample. In contrast, data that is not generated by the trained model LM may be referred to as a real sample. As an example, a real sample is a sample based on data input by the user or another worker. The above names do not limit this exemplary embodiment.

[0097] (Specific example of learning process by learning unit 23 and conversion unit 13) A specific example of the learning process performed by the learning unit 23 and the conversion unit 13 will be described below.

[0098] (Problem setting) First, we will explain the problem setting (various definitions) in this example. The set of users is denoted as U, A set of items is denoted as I, The interaction score (sometimes called the correlation score) between user u (u is an element of set U) and item i (i is an element of set I) is expressed as r u,i It is written as follows.

[0099] where r u,i is an element of the set R of observed interactions, where R is

number

number

number

[0100] The text in question u can be regarded as an example of the above-mentioned user attribute information. i can be regarded as an example of the above-mentioned item attribute information. u and t i As the training data, data (real samples) that are not generated by the trained model LM are used.

[0101] These texts u ,t i For example, the vector representation is obtained by the TF-IDF (Term Frequency - Inverse Document Frequency) algorithm or the BERT algorithm executed by the conversion unit 13.

number

[0102] (Based framework) Next, we will explain the framework that is the premise of the conversion process executed by the conversion unit 13. The framework described below can also be understood as the so-called Heater algorithm described in Non-Patent Document 1, for example, but this term does not limit this exemplary embodiment.

[0103] In this example, as partially described above, a user u and an item i are converted into vectors f(u) and f'(i) by the conversion unit 13. Here, f(u) and f'(i) are vectors in an l-dimensional space. For such conversion into vectors, a configuration employed in a recommendation system, typified by Matrix Factorization (MF), can be used, for example, but this example does not limit this exemplary embodiment.

[0104] Then, the prediction unit 14 calculates the predicted value ^r of the similarity. u,i (R with a hat u,i )of

number

[0105] In Matrix Factorization, for example, the observed interactions R are used as training data to generate latent representations for user u and item i.

number

[0106] On the other hand, in the processing of the conversion unit 13 and the like in the framework underlying this example, embedding derived from side information is also a target of learning. This can be considered a process similar to collaborative filtering embedding, but this term does not limit this exemplary embodiment.

[0107] In this framework, first, the user's side information x u But in the collaborative filtering embedding space,

number

number

[0108] In the optimization process in the underlying framework, the parameters

number

number

number

[0109] Once the above learning is complete, for warm-start user u and cold-start item i, <Vpu,V’W’x u > is the actual reaction (interaction) r u,i It is expected that the inference can be performed using only side information without using collaborative filtering. It is also important to note that the above process is performed by inner product and matrix multiplication calculations and does not involve costly processing.

[0110] (Learning process by the learning unit 23 and the conversion unit 13 according to this example) The learning unit 23 and the conversion unit 13 according to this example perform processing that takes into consideration not only the above-described premise framework but also the contribution of pseudo samples (data generated by the trained model LM). More specifically, the learning unit 23 updates the parameters used by the conversion unit 13 in the conversion processing so as to minimize the following loss function:

number

[0111] In this way, the learning unit 23 controls the conversion unit 13 as follows: A first loss value (for example, the above L) obtained by referring to the data (pseudo samples) generated by the trained model LM aug )and, A second loss value (e.g., L) obtained by referring to data (real samples) other than the data generated by the trained model. heater )and, The loss function (L stove ) to train it.

[0112] (Example of pseudo sample generation by the learning unit 23) The following describes an example of generating pseudo samples used by the learning unit 23 to train the conversion unit 13. As partially described above, the learning unit 23 uses data (pseudo samples) generated by the trained model LM, as an example, to train the conversion unit 13.

[0113] As an example, the learning unit 23 Generate one or more prompts PR by referring to the information (at least one of information about the user and information about the item) acquired by the acquisition unit 11 (12, 21, 22), Inputting the generated one or more prompts PR into the trained model LM via the communication unit 30; Generate the above pseudo-samples based on the data output by the trained model LM (answers by the trained model LM). The following process is performed.

[0114] Below, an example of a prompt PR generated by the learning unit 23, a response to the prompt PR using the trained model LM, and a pseudo sample generated based on the response will be described.

[0115] (Example 1) 9 is a diagram showing example 1 of a prompt PR generated by the learning unit 23, an answer of the trained model LM, and a pseudo sample. As shown in FIG. 9, the prompt PR1 generated by the learning unit 23 is -Text indicating that there is an item (measure) under consideration (PR11 in Figure 9) -Text indicating the content of the item under consideration (PR_ITEM1) -Text indicating user presence (PR12) One or more texts that describe the user (PR_USER1 and PR_USER2) -Text asking whether the user will respond to the item (measure) (PR13) Contains:

[0116] In the example shown in Figure 9, the trained model LM to which the prompt PR1 is input is A response indicating that user A responds to the item and user B does not respond to the item (RES1) and provides it to the learning unit 23. Then, the learning unit 23 refers to the response RES1 and obtains At least one of the ID and text of the item (measure) -Text (positive example) of user A who responds to the item (measure), -Text of User B (negative example) who does not respond to the item (measure) The pseudo sample PD1 thus generated is, as described above, for example, a loss function L aug and is used for training of the conversion unit 13 by the training unit 23.

[0117] (Example 2) 10 is a diagram showing example 2 of a prompt PR generated by the learning unit 23, an answer of the trained model LM, and a pseudo sample. As shown in FIG. 10, the prompt PR2 generated by the learning unit 23 is -Text indicating that there is an item (measure) under consideration (PR21 in Figure 10) -Text indicating the content of the item under consideration (PR_ITEM1) Text inquiring about the characteristics of users who respond well to the above items (measures) and the characteristics of users who do not respond to the above items (measures) (PR22) Contains:

[0118] In the example shown in Figure 10, the answer RES2 generated by the trained model LM to which the above prompt PR2 is input is, for example, a text that expresses the characteristics of a user who responds favorably. People who are interested in foreign cuisine (Responding User 1) People who enjoy cooking (Responding User 2) People looking for a fresh and unique dining experience (Responding User 3), And text that expresses the characteristics of unresponsive users People who don't care about the food and just want something that will fill them up (User 1 who didn't respond) People with limited budgets (non-responders 2) People who are reluctant to use spices in cooking (Non-Respondent User 3) The present invention is configured to include the following.

[0119] The learning unit 23 refers to the response RES2 and At least one of the ID and text of the item (measure) -Text (positive example) of the above-mentioned user 1 who reacts to the item (measure), and the text (negative example) of the non-responding user 1 who is a user who does not respond to the item (measure). The learning unit 23 also generates pseudo samples PD22, PD23, ... by combining the item (measure) with texts of other users. In this example, the learning unit 23 can generate a pseudo sample group PD2 including 3 x 3 = 9 pseudo samples. The pseudo sample PD2 generated in this way is calculated using, as an example, the loss function L aug and is used for training of the conversion unit 13 by the training unit 23.

[0120] As in this example, by using prompts to query the trained model LM about multiple user features, pseudo-samples containing a large number of combinations can be generated, allowing the learning unit 23 to more efficiently train the conversion unit 13.

[0121] (Example 3) 11 is a diagram showing Example 3 of a prompt PR generated by the learning unit 23, an answer of the trained model LM, and a pseudo sample. As shown in FIG. 11, the prompt PR3 generated by the learning unit 23 is - Text indicating that an item (measure) has been executed in the past (PR31 in Figure 11) -Text showing the contents of the item (measure) executed in the past (PR_ITEM1) -Texts explaining the user's reaction at that time (PR32, PR_USER1, PR_USER2, PR_USER3) -Text indicating that there is an item (measure) under consideration (PR33) -Text indicating the content of the item under consideration (PR_ITEM2) Text indicating the presence of a user and text describing that user (PR34) -Text asking whether the user will respond to the item (measure) (PR35) Contains:

[0122] In the example shown in FIG. 11, the trained model LM to which the prompt PR3 is input is A response indicating that user A responds to the item and user B does not respond to the item (RES3) and provides it to the learning unit 23. Then, the learning unit 23 refers to the response RES3 and obtains At least one of the ID and text of the item (measure) -Text (positive example) of user A who responds to the item (measure), -Text of User B (negative example) who does not respond to the item (measure) The pseudo sample PD1 thus generated is, as described above, for example, a pseudo sample PD3 including the loss function L aug and is used for training of the conversion unit 13 by the training unit 23.

[0123] As in this example, by including in the prompt the items (measures) that have been implemented in the past and the user's reaction at that time, a more suitable answer can be obtained from the trained model LM, and the learning unit 23 can generate a more suitable pseudo-sample PD.

[0124] (Example 4) 12 is a diagram showing Example 4 regarding a prompt PR generated by the learning unit 23, an answer of the trained model LM, and a pseudo sample. As shown in FIG. 12, the prompt PR4 generated by the learning unit 23 is Rating history (PR41) of one or more items (Movie A to Movie E in the example of FIG. 12) by a certain user (a user with user ID 001 in the example of FIG. 12) -Text (PR42) inquiring about the characteristics of the item (movie) that the user would like based on the rating history above In other words, the prompt PR4 according to this example includes a query statement (PR42) that inquires about information about an item based on the user information acquired by the acquisition unit 11 (12, 21, 22).

[0125] In the example shown in FIG. 12, the trained model LM to which the prompt PR4 is input is Responses (RES4) including the features of the item (movie) that the user would like and the user's predicted evaluation of those features ·“animated with fantasy elements”: 5.0 ·“action-packed superhero film”: 5.0 ·“classic fairytale adaptation”: 5.0 ·“sci-fi with strong visuals”: ​​5.0 and provides it to the learning unit 23. Then, the learning unit 23 refers to the response RES4 and obtains At least one of the user's ID and text ·The characteristics of the movie that the user likes and The pseudo sample PD4 thus generated is, as described above, for example, a loss function L aug and is used for training of the conversion unit 13 by the training unit 23.

[0126] The prompt PR4 in this example includes a query (PR42) that inquires about information about an item based on the user information acquired by the acquisition unit 11 (12, 21, 22), so that a useful answer can be obtained from the trained model LM, allowing the training unit 23 to generate a more suitable pseudo sample PD.

[0127] As described above, the learning unit 23 Acquires data (pseudo samples) generated by the trained model LM by referring to information (at least one of information about the user and information about the item) acquired by the first acquisition unit 11 (12, 21, 22), One or more conversion units 13 that convert each of the users and the items into vectors are trained using training data including data (pseudo samples) generated by the trained model LM. This configuration allows for an increase in the amount of training data. Furthermore, the conversion unit 13 and prediction unit 14, which have been trained using such training data, perform prediction processing using vector transformation, eliminating the need to access the trained model LM during the inference phase. This makes it possible to provide a recommendation technology that can address the cold start problem while suppressing increases in costs.

[0128] (Calculation process of threshold value referred to by prediction unit 14) The calculation process of the threshold value referred to by the prediction unit 14 will be described below. The learning unit 23 sets (derives) a threshold value by referring to the similarity of data included in the training data to which the learning unit 23 refers, The above-mentioned prediction process is performed by referring to the set (derived) threshold value. As an example, the prediction unit 14 may refer to one or more pairs used by the learning unit 23 in the learning phase, and may Similarity between the item (measure) and the positive example for that item (first similarity), Similarity between the item (measure) and the negative example for that item (second similarity) may be calculated, and an intermediate value between the first similarity and the second similarity may be set as the threshold value.

[0129] For example, the prediction unit 14 Similarity between the policy (item) and the positive example pair (first similarity: S_pos), Similarity between the pair of measures (items) and negative examples (second similarity: S_neg) Calculate the mean of the similarity of the positive examples (mean(S_pos)) and the mean of the similarity of the negative examples (mean(S_neg)), threshold = ( mean(S_pos) + mean(s_neg) ) / 2 The threshold may be derived by:

[0130] Alternatively, the prediction unit 14 The threshold value may be calculated by averaging a negative example whose similarity exceeds the lowest similarity among the positive examples being trained and a positive example whose similarity is less than the highest similarity among the negative examples being trained.

number

[0131] Third Embodiment A second exemplary embodiment, which is one example of an embodiment of the present invention, will be described in detail with reference to the drawings. Components having the same functions as those described in the above exemplary embodiment will be assigned the same reference numerals, and their description will be omitted as appropriate. The scope of application of each technology employed in this exemplary embodiment is not limited to this exemplary embodiment. That is, each technology employed in this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs. Furthermore, each technology shown in each drawing referenced to explain this exemplary embodiment can also be employed in other exemplary embodiments included in the present disclosure, to the extent that no particular technical hindrance occurs.

[0132] (Configuration of information processing device 200) The configuration of the information processing device 200 according to this exemplary embodiment will be described with reference to Fig. 13. As shown in Fig. 13, the information processing device 200 does not include the learning unit 23, which is one of the components included in the information processing device 100 according to exemplary embodiment 2. The other components are the same as those of the information processing device 100.

[0133] Moreover, the information processing device 200 according to the present exemplary embodiment includes, as an example, a conversion unit 13 that has been trained by the learning unit 23 included in the information processing device 100 according to exemplary embodiment 2. In this way, the information processing device 200 A first acquisition means (acquisition unit 11 (12, 21, 22)) for acquiring information about a user; A second acquisition means (acquisition unit 11 (12, 21, 22)) for acquiring information about an item; One or more conversion means (conversion unit 13) that are trained using training data including data generated by a trained model and convert each of the information about the user and the information about the item into a vector; a prediction means (prediction unit 14) for predicting the compatibility between the user and the item by referring to the vector converted by the conversion means; This configuration also provides the various effects of the exemplary embodiments described above.

[0134] (Additional Notes Regarding Each Embodiment) In the above description, an example has been given in which the compatibility between a user and an item is learned and predicted, but this does not limit the matters described in this specification. A first acquisition means for acquiring information about a first subject (for example, a user); A second acquisition means for acquiring information about a second object (for example, an item); one or more conversion means that are trained using training data including data generated by a trained model and convert each of the information about the first object and the information about the second object into a vector; a prediction means for predicting the compatibility between the first target and the second target by referring to the vector converted by the conversion means; It may also be expressed as an information processing device equipped with the above.

[0135] In addition, the information processing device 100 a first acquisition means for acquiring at least one of information relating to a first object (for example, a user) and information relating to a second object (for example, an item); a second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means; a learning means for learning one or more conversion means for converting each of the first object and the second object into a vector using training data including the data acquired by the second acquisition means; It may also be expressed as an information processing device equipped with the above.

[0136] With the above configuration, it is possible to provide a recommendation technology that can deal with the cold start problem while suppressing increases in costs.

[0137] [Software implementation example] Some or all of the functions of the information processing devices 1, 2, 100, and 200 (hereinafter also referred to as "each of the above devices") may be realized by hardware such as an integrated circuit (IC chip), or by software.

[0138] In the latter case, each of the above devices is realized by, for example, a computer that executes instructions of a program, which is software that realizes each function. An example of such a computer (hereinafter referred to as computer C) is shown in Figure 14. Figure 14 is a block diagram showing the hardware configuration of computer C that functions as each of the above devices.

[0139] The computer C includes at least one processor C1 and at least one memory C2. The memory C2 stores a program P for causing the computer C to operate as each of the above-mentioned devices. In the computer C, the processor C1 reads and executes the program P from the memory C2, thereby realizing the functions of each of the above-mentioned devices.

[0140] The processor C1 may be, for example, a central processing unit (CPU), a graphic processing unit (GPU), a digital signal processor (DSP), a micro processing unit (MPU), a floating point number processing unit (FPU), a physics processing unit (PPU), a tensor processing unit (TPU), a quantum processor, a microcontroller, or a combination thereof. The memory C2 may be, for example, a flash memory, a hard disk drive (HDD), a solid state drive (SSD), or a combination thereof.

[0141] The computer C may further include a RAM (Random Access Memory) for expanding the program P during execution and for temporarily storing various data. The computer C may also include a communication interface for transmitting and receiving data to and from other devices. The computer C may also include an input / output interface for connecting input / output devices such as a keyboard, mouse, display, and printer.

[0142] Furthermore, the program P can be recorded on a non-transitory tangible recording medium M that can be read by the computer C. Such a recording medium M can be, for example, a tape, a disk, a card, a semiconductor memory, or a programmable logic circuit. The computer C can acquire the program P via such a recording medium M. The program P can also be transmitted via a transmission medium. Such a transmission medium can be, for example, a communication network or broadcast waves. The computer C can also acquire the program P via such a transmission medium.

[0143] [Appendix A] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0144] (Appendix A1) a first acquiring means for acquiring information about a user; a second acquiring means for acquiring information about the item; One or more conversion means that are trained using training data including data generated by a trained model and convert each of the information about the user and the information about the item into a vector; a prediction means for predicting compatibility between the user and the item by referring to the vector converted by the conversion means; An information processing device comprising:

[0145] (Appendix A2) The items include at least one of a product and a measure. 10. The information processing device according to claim 1,

[0146] (Appendix A3) The one or more conversion means a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and It is trained using a loss function in which different weights are applied to An information processing device according to appendix A1 or A2.

[0147] (Appendix A4) The prediction means A threshold value is set by referring to the similarity of data included in the training data, and a compatibility between the user and the item is predicted. An information processing device according to any one of appendices A1 to A3.

[0148] (Appendix A5) The prediction means Deriving a similarity between the user vector generated by the one or more conversion means and the item vector; The derived similarity is referenced to predict the compatibility between the user and the item. An information processing device according to any one of appendices A1 to A4.

[0149] (Appendix A6) The one or more transformation means are trained by contrastive learning. An information processing device according to any one of appendices A1 to A5.

[0150] (Appendix A7) a first acquisition means for acquiring at least one of information about a user and information about an item; a second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means; a learning means for learning one or more conversion means for converting each of the users and the items into vectors using training data including the data acquired by the second acquisition means; An information processing device comprising:

[0151] (Appendix A8) The second acquisition means generating one or more prompts by referring to the information acquired by the first acquisition means; The data output by the trained model to which the generated one or more prompts have been input is acquired as data generated by the trained model. 10. The information processing device according to claim 7,

[0152] (Appendix A9) The prompt may include: The first acquisition means includes a query statement for inquiring about information about an item based on the user information acquired by the first acquisition means. 10. The information processing device according to claim 8,

[0153] (Appendix A10) The learning means may be configured to: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and Train using a loss function that applies different weights to 10. The information processing device according to claim 9, wherein the information processing device is a device for processing information.

[0154] (Appendix A11) The learning means training said one or more transforming means by contrastive learning; 10. The information processing device according to claim 7, wherein the information processing device is a device for processing information.

[0155] [Appendix B] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0156] (Appendix B1) a first acquisition process in which at least one processor acquires information about a user; a second acquisition process in which the at least one processor acquires information about an item; a transformation process in which the at least one processor transforms each of the information about the user and the information about the item into a vector using one or more transformations learned using training data including data generated by a trained model; a prediction process in which the at least one processor predicts compatibility between the user and the item by referring to the vector converted by the conversion process; An information processing method comprising:

[0157] (Appendix B2) The items include at least one of a product and a measure. 1. The information processing method described in Appendix B1.

[0158] (Appendix B3) The one or more conversion processes are: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and It is trained using a loss function in which different weights are applied to 1. An information processing method according to Appendix B1 or B2.

[0159] (Appendix B4) In the prediction process, the at least one processor: A threshold value is set by referring to the similarity of data included in the training data, and a compatibility between the user and the item is predicted. 10. The information processing method according to any one of appendices B1 to B3.

[0160] (Appendix B5) In the prediction process, the at least one processor: Deriving a similarity between the user vector generated by the one or more conversion processes and the item vector; The derived similarity is referenced to predict the compatibility between the user and the item. 1. An information processing method according to any one of appendices B1 to B4.

[0161] (Appendix B6) The one or more transformation processes are learned by contrastive learning. 1. An information processing method according to any one of Appendices B1 to B5.

[0162] (Appendix B7) a first acquisition process in which the at least one processor acquires at least one of information about a user and information about an item; A second acquisition process in which the at least one processor acquires data generated by a trained model by referring to the information acquired by the first acquisition process; a learning process in which the at least one processor trains one or more conversion means that convert each of the users and the items into vectors using training data including the data acquired by the second acquisition process; An information processing method comprising:

[0163] (Appendix B8) The second acquisition process includes: generating one or more prompts by referring to the information acquired by the first acquisition process; The data output by the trained model to which the generated one or more prompts have been input is acquired as data generated by the trained model. An information processing method as described in Appendix B7.

[0164] (Appendix B9) The prompt may include: The first acquisition process includes a query statement for inquiring about information about an item based on the acquired user information. The information processing method described in Appendix B8.

[0165] (Appendix B10) The at least one processor, in the learning process, performs the one or more conversion processes as follows: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and Train using a loss function that applies different weights to An information processing method according to any one of Appendices B7 to B9.

[0166] (Appendix B11) In the learning process, The at least one processor trains the one or more transformation processes by contrastive learning. 10. An information processing method according to any one of appendices B7 to B10.

[0167] [Appendix C] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0168] (Appendix C1) A program that causes a computer to function as an information processing device, The computer a first acquiring means for acquiring information about a user; a second acquiring means for acquiring information about the item; One or more conversion means that are trained using training data including data generated by a trained model and convert each of the information about the user and the information about the item into a vector; a prediction means for predicting compatibility between the user and the item by referring to the vector converted by the conversion means; An information processing program that functions as a

[0169] (Appendix C2) The items include at least one of a product and a measure. An information processing program as described in Appendix C1.

[0170] (Appendix C3) The one or more conversion means a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and It is trained using a loss function in which different weights are applied to An information processing program according to appendix C1 or C2.

[0171] (Appendix C4) The prediction means A threshold value is set by referring to the similarity of data included in the training data, and a compatibility between the user and the item is predicted. An information processing program according to any one of appendices C1 to C3.

[0172] (Appendix C5) The prediction means Deriving a similarity between the user vector generated by the one or more conversion means and the item vector; The derived similarity is referenced to predict the compatibility between the user and the item. An information processing program according to any one of appendices C1 to C4.

[0173] (Appendix C6) The one or more transformation means are trained by contrastive learning. An information processing program according to any one of appendices C1 to C5.

[0174] (Appendix C7) The computer a first acquisition means for acquiring at least one of information about a user and information about an item; a second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means; a learning process in which one or more conversion means for converting each of the users and the items into vectors is trained using training data including the data acquired by the second acquisition means; An information processing program that functions as a

[0175] (Appendix C8) The second acquisition means generating one or more prompts by referring to the information acquired by the first acquisition means; The data output by the trained model to which the generated one or more prompts have been input is acquired as data generated by the trained model. An information processing program as described in Appendix C7.

[0176] (Appendix C9) The prompt may include: The first acquisition means includes a query statement for inquiring about information about an item based on the user information acquired by the first acquisition means. An information processing program as described in Appendix C8.

[0177] (Appendix C10) The learning means may be configured to: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and Train using a loss function that applies different weights to An information processing program according to any one of appendices C7 to C9.

[0178] (Appendix C11) The learning means training said one or more transforming means by contrastive learning; An information processing program according to any one of appendices C7 to C10.

[0179] [Appendix D] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0180] (Appendix D1) at least one processor, a first acquisition process for acquiring information about a user; a second acquisition process for acquiring information about the item; one or more conversion processes for converting each of the information about the user and the information about the item into a vector using one or more conversion means trained using training data including data generated by a trained model; a prediction process for predicting compatibility between the user and the item by referring to the vector converted by the conversion process; An information processing device that executes the above.

[0181] The information processing device may further include a memory, and the memory may store a program for causing the at least one processor to execute each of the processes.

[0182] (Appendix D2) The items include at least one of a product and a measure. 10. The information processing device according to claim 9, wherein the information processing device is a device for processing information.

[0183] (Appendix D3) The one or more conversion processes are: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and It is trained using a loss function in which different weights are applied to An information processing device according to appendix D1 or D2.

[0184] (Appendix D4) In the prediction process, the at least one processor: A threshold value is set by referring to the similarity of data included in the training data, and a compatibility between the user and the item is predicted. An information processing device according to any one of appendices D1 to D3.

[0185] (Appendix D5) In the prediction process, the at least one processor: Deriving a similarity between the user vector generated by the one or more conversion processes and the item vector; The derived similarity is referenced to predict the compatibility between the user and the item. An information processing device according to any one of appendices D1 to D4.

[0186] (Appendix D6) The one or more transformation processes are learned by contrastive learning. An information processing device according to any one of appendices D1 to D5.

[0187] (Appendix D7) The at least one processor: a first acquisition process for acquiring at least one of information about a user and information about an item; A second acquisition process that acquires data generated by the trained model by referring to the information acquired by the first acquisition process; a learning process that learns one or more conversion processes that convert each of the users and the items into vectors using training data including the data acquired by the second acquisition process; An information processing device that executes the above.

[0188] (Appendix D8) The second acquisition process includes: generating one or more prompts by referring to the information acquired by the first acquisition process; The data output by the trained model to which the generated one or more prompts have been input is acquired as data generated by the trained model. 10. The information processing device according to claim 7,

[0189] (Appendix D9) The prompt may include: The first acquisition process includes a query statement for inquiring about information about an item based on the acquired user information. 10. The information processing device according to claim 8,

[0190] (Appendix D10) In the learning process, the at least one processor performs the one or more conversion processes as follows: a first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and Train using a loss function that applies different weights to An information processing device according to any one of appendices D7 to D9.

[0191] (Appendix D11) In the learning process, the at least one processor The one or more transformation processes are trained by contrastive learning. An information processing device according to any one of appendices D7 to D10.

[0192] [Appendix E] This disclosure includes the techniques described in the following appendices. However, the present invention is not limited to the techniques described in the following appendices, and various modifications are possible within the scope of the claims.

[0193] (Appendix E1) A program that causes a computer to function as an information processing device, The computer, a first acquisition process for acquiring information about a user; a second acquisition process for acquiring information about the item; One or more transformation processes that are trained using training data including data generated by a trained model and that transform each of the information about the user and the information about the item into a vector; a prediction process for predicting compatibility between the user and the item by referring to the vector converted by the conversion process; A non-transitory recording medium on which an information processing program for executing the above is recorded. [Explanation of symbols]

[0194] 1, 2, 100, 200 ··· Information processing device 11, 21 First acquisition unit (first acquisition means) 12, 22 Second acquisition unit (second acquisition means) 13. Conversion unit (conversion means) 14. Prediction unit (prediction means) 23 Learning section (learning means)

Claims

1. a first acquiring means for acquiring information about a user; a second acquisition means for acquiring information about the item; One or more conversion means that are trained using training data including data generated by a trained model and that convert each of the information about the user and the information about the item into a vector; a prediction means for predicting compatibility between the user and the item by referring to the vector converted by the conversion means; An information processing device comprising:

2. The items include at least one of a product and a measure. The information processing device according to claim 1 .

3. The one or more conversion means A first loss value obtained by referring to data generated by the trained model; a second loss value obtained by referring to data other than the data generated by the trained model; and It is trained using a loss function in which different weights are applied to The information processing device according to claim 2 .

4. a first acquisition means for acquiring at least one of information about a user and information about an item; A second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means; a learning means for learning one or more conversion means for converting each of the users and the items into vectors using training data including the data acquired by the second acquisition means; An information processing device comprising:

5. The second acquisition means generating one or more prompts by referring to the information acquired by the first acquisition means; The data output by the trained model to which the generated one or more prompts have been input is acquired as data generated by the trained model. The information processing device according to claim 4 .

6. The prompt may include: The first acquisition means includes a query statement for inquiring about information about an item based on the user information acquired by the first acquisition means. The information processing device according to claim 5 .

7. Obtaining information about a user; Obtaining information about the item; converting each of the information about the user and the information about the item into a vector by one or more conversion means trained using training data including data generated by a trained model; predicting compatibility between the user and the item by referring to the transformed vector; An information processing method comprising:

8. Obtaining at least one of information about a user and information about an item; acquiring data generated by the trained model by referring to the acquired information; training one or more conversion means for converting each of the users and the items into vectors using training data including the acquired data; An information processing method comprising:

9. A program that causes a computer to function as an information processing device, The computer a first acquiring means for acquiring information about a user; a second acquisition means for acquiring information about the item; One or more conversion means that are trained using training data including data generated by a trained model and that convert each of the information about the user and the information about the item into a vector; a prediction means for predicting compatibility between the user and the item by referring to the vector converted by the conversion means; A program that functions as a

10. A program that causes a computer to function as an information processing device, The computer a first acquisition means for acquiring at least one of information about a user and information about an item; A second acquisition means for acquiring data generated by a trained model by referring to the information acquired by the first acquisition means; a learning means for learning one or more conversion means for converting each of the users and the items into vectors using training data including the data acquired by the second acquisition means; A program that functions as a

Images