Information processing device, information processing method, and recording medium

Abstract

This information processing device comprises: a compression means that generates user information obtained by compressing past record information indicating past records of a user; a creation means that creates a prediction prompt for predicting a recommended item to be recommended to the user, the prediction prompt including the user information; and a prediction means that uses the prediction prompt to predict the recommended item using a large language model (LLM) and outputs the recommended item.

Description

Information processing device, information processing method, and recording medium

[0001] The present disclosure relates to the technical fields of an information processing device, an information processing method, and a recording medium.

[0002] As a technique for predicting recommended products, a technique has been proposed in which a prompt is input into a large language model, and as a result, a recommended product category and the corresponding recommended product are obtained (see Patent Document 1).

[0003] Japanese Patent Application Laid-Open No. 2023-129333

[0004] An object of this disclosure is to provide an information processing device, an information processing method, and a recording medium that aim to improve the technology related to the prior art documents mentioned above.

[0005] One aspect of an information processing device includes a compression means for generating user information by compressing past performance information indicating a user's past performance, a creation means for creating a predictive prompt including the user information and for predicting recommended items to be recommended to the user, and a prediction means for predicting the recommended items using a large language model (LLM) using the predictive prompt and outputting the recommended items.

[0006] One aspect of the information processing method is an information processing method executed by a computer, which includes generating user information by compressing past performance information indicating the user's past performance, creating a predictive prompt that includes the user information and predicts recommended items to be recommended to the user, and using the predictive prompt to predict the recommended items using Large Language Models (LLMs) and output the recommended items.

[0007] One aspect of the recording medium has recorded thereon a computer program for causing a computer to execute an information processing method, which includes generating user information by compressing past performance information indicating a user's past performance, creating a predictive prompt that includes the user information and predicts recommended items to be recommended to the user, and using the predictive prompt to predict the recommended items using Large Language Models (LLMs) and output the recommended items.

[0008] 1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment. FIG. 1 is a flowchart showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment. FIG. 1 is a flowchart showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a block diagram showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a schematic diagram showing an example of an outline of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment. FIG. 1 is a flowchart showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment. FIG. 1 is a flowchart showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a block diagram showing an example of a configuration of an information processing device according to an embodiment. FIG. 1 is a block diagram showing an example of a flow of information processing operation in an information processing device according to the present disclosure. FIG. 1 is a schematic diagram showing an example of an outline of information processing operation in an information processing device according to the present disclosure.

[0009] Hereinafter, an information processing device, an information processing method, and a recording medium according to an embodiment will be described with reference to the drawings. [1: First Embodiment]

[0010] A first embodiment of an information processing device, an information processing method, and a recording medium will be described with reference to Fig. 1 and Fig. 2. In the following, the first embodiment of the information processing device, the information processing method, and the recording medium will be described using an information processing device 10.

[0011] 1, the information processing device 10 includes a compression unit 11, a creation unit 12, and a prediction unit 13. The operation of the information processing device 10 will be described with reference to the flowchart of FIG.

[0012] 2, the compression unit 11 generates user information by compressing information indicating the user's past performance (step S11). The user information is information obtained by compressing information indicating the user's past performance (referred to as "past performance information"). Compression is an operation of processing or converting information to reduce the volume of information.

[0013] A user may be an individual or a corporation. A user may, for example, be a user of a service provided by a service provider. The service provider may, for example, be an e-commerce site. The information indicating the user's past performance may be information indicating the user's service usage history. The user's service usage history may include information indicating the user's reaction to an item. The reaction may be, for example, information indicating whether the item was purchased, a score given on a review site, the number of times the video was viewed if the item is a video, or the number of times the song was listened to if the item is a song.

[0014] The user's service usage history may be the user's item purchase history. In this case, the past performance information may be information indicating items purchased by the user. Items purchased by the user in the past may be referred to as past performance items. The past performance information may also be referred to as a group of past performance items. The user's past performance may also be campaigns that the user has previously implemented and the results of those campaigns. In this case, the user may be a corporation.

[0015] The creation unit 12 creates a predictive prompt for predicting a recommended item to be recommended to the user (step S12). The recommended item to be recommended to the user may be an item predicted to be desired by the user for purchase. The recommended item to be recommended to the user may be an item predicted to be desired by the user for viewing. The recommended item to be recommended to the user may be a campaign predicted to be desired by the user. The predictive prompt created by the creation unit 12 includes the user information generated by the compression unit 11. That is, the predictive prompt created by the creation unit 12 includes compressed information on the user's past performance information. Because the user information is compressed information, the predictive prompt created by the creation unit 12 has a smaller capacity than a predictive prompt including uncompressed information on the user's past performance information.

[0016] The prediction unit 13 predicts recommended items to be recommended to the user (step S13). The prediction unit 13 predicts recommended items using Large Language Models (LLMs) by using the predicted prompts created by the creation unit 12. Since the predicted prompts include compressed information on the user's past performance, the prediction unit 13 predicts recommended items based on the user's past performance.

[0017] In this way, the information processing device 10 performs an information processing method that includes generating user information by compressing past performance information that indicates the user's past performance, creating a predictive prompt that includes the user information and predicts recommended items to be recommended to the user, predicting recommended items using the predictive prompt with LLM, and outputting the recommended items.

[0018] The information processing device 10 described above may be realized by a computer reading a computer program recorded on a recording medium. In this case, the computer program may cause the computer to execute an information processing method including: generating user information by compressing past performance information indicating the user's past performance; creating a predictive prompt including the user information and for predicting recommended items to be recommended to the user; predicting recommended items using the predictive prompt with an LLM; and outputting the recommended items. [Technical Effect]

[0019] The information processing device 10 according to the present disclosure generates a prompt including compressed information on past performance information, and therefore can perform predictions using small-capacity predictive prompts. This reduces the prediction time and the processing load of predictions. [2: Second Embodiment]

[0020] A second embodiment of the information processing device, information processing method, and recording medium will be described with reference to FIGS. 3 to 6. Hereinafter, the second embodiment of the information processing device, information processing method, and recording medium will be described using an information processing device 20. Note that, for the second embodiment, descriptions that overlap with the description of the first embodiment will be omitted as appropriate. [2-1: Configuration of the information processing device 20]

[0021] The configuration of the information processing device 20 will be described with reference to Fig. 3. Fig. 3 is a block diagram showing the configuration of the information processing device 20.

[0022] 3 , the information processing device 20 includes a calculation device 21 and a storage device 22. The information processing device 20 may further include a communication device 23, an input device 24, and an output device 25. However, the information processing device 20 does not necessarily include at least one of the communication device 23, the input device 24, and the output device 25. The calculation device 21, the storage device 22, the communication device 23, the input device 24, and the output device 25 may be connected via a data bus 26.

[0023] The arithmetic device 21 includes at least one processor (i.e., one processor or multiple processors) as hardware. The processor may include, for example, a processor conforming to a von Neumann computer architecture. The processor conforming to the von Neumann computer architecture may include at least one of a CPU (Central Processing Unit) and a GPU (Graphics Processing Unit). The processor may include, for example, a processor conforming to a non-von Neumann computer architecture. The processor conforming to the non-von Neumann computer architecture may include at least one of an FPGA (Field Programmable Gate Array) and an ASIC (Application Specific Circuit).

[0024] The arithmetic device 21 reads a computer program 221 including at least one of computer program code and computer program instructions. For example, the arithmetic device 21 may read the computer program 221 stored in the storage device 22. For example, the arithmetic device 21 may read the computer program 221 stored in a computer-readable, non-transitory recording medium using a recording medium reading device (not shown) included in the information processing device 20. The computer program 221 read from the recording medium may be stored in the storage device 22. The arithmetic device 21 may acquire (i.e., download or read) the computer program 221 from a device (not shown) located outside the information processing device 20 via the communication device 23 (or another communication device). The downloaded computer program 221 may be stored in the storage device 22.

[0025] The arithmetic device 21 executes the loaded computer program 221. As a result, logical functional blocks for executing the information processing method to be performed by the information processing device 20 are realized within the arithmetic device 21. In other words, the arithmetic device 21, together with the storage device 22 or the like in which the computer program 221 is recorded (in other words, together with the storage device 22 and the computer program 221 recorded in the storage device 22 or the like), can function as a controller or computer for realizing logical functional blocks for executing the processing to be performed by the information processing device 20. In other words, the at least one processor included in the arithmetic device 21, the memory (recording medium) included in the storage device 22 or the like, and the computer program 221 are configured so that the information processing device 20 performs the information processing to be performed by the information processing device 20.

[0026] A computational model that can be constructed by machine learning may be implemented in the computational device 21 by the computational device executing the computer program 221. An example of a computational model that can be constructed by machine learning is a computational model including a neural network (so-called artificial intelligence (AI)). In this case, learning of the computational model may include learning of parameters of the neural network (e.g., at least one of a weight and a bias). The computational device 21 may execute an information processing method using the computational model. In other words, the operation of executing the information processing method may include the operation of executing the information processing method using the computational model. Note that a computational model that has been constructed by offline machine learning using training data may be implemented in the computational device 21. Furthermore, the computational model implemented in the computational device 21 may be updated by online machine learning on the computational device 21. Alternatively, the calculation device 21 may execute the information processing method using a calculation model implemented in a device external to the calculation device 21 (i.e., a device provided outside the information processing device 20) in addition to or instead of the calculation model implemented in the calculation device 21.

[0027] The recording medium for recording the computer program 221 executed by the arithmetic device 21 may be at least one of a CD-ROM, CD-R, CD-RW, flexible disk, MO, DVD-ROM, DVD-RAM, DVD-R, DVD+R, DVD-RW, DVD+RW, Blu-ray (registered trademark), or other optical disk, a magnetic medium such as a magnetic tape, a magneto-optical disk, a semiconductor memory such as a USB memory, or any other medium capable of storing a program. The recording medium may include a device capable of recording a computer program (for example, a general-purpose device or a dedicated device in which the computer program 221 is implemented in a state in which it can be executed in at least one of the forms of software and firmware). Furthermore, each process or function included in the computer program 221 may be realized by a logical processing block realized within the arithmetic device 21 when the arithmetic device 21 (i.e., processor) executes the computer program 221, or may be realized by hardware such as a predetermined gate array (FPGA (Field Programmable Gate Array), ASIC (Application Specific Integrated Circuit)) provided in the arithmetic device 21, or may be realized in a form that mixes logical processing blocks and partial hardware modules that realize some elements of the hardware.

[0028] The storage device 22 includes at least one memory capable of storing desired data. In other words, the storage device 22 includes at least one memory containing desired data. For example, the storage device 22 may store a computer program 221 executed by the arithmetic device 21. In this case, the storage device 22 (memory) may be used as the above-mentioned recording medium for recording the computer program 221 executed by the arithmetic device 21. The storage device 22 may temporarily store data used by the arithmetic device 21 when the arithmetic device 21 is executing the computer program 221. The storage device 22 may also store data to be stored long-term by the information processing device 20. The storage device 22 may include at least one of a RAM (Random Access Memory), a ROM (Read Only Memory), a hard disk device, a magneto-optical disk device, an SSD (Solid State Drive), and a disk array device. In other words, the storage device 22 may include a non-temporary recording medium.

[0029] The communication device 23 may be capable of communicating with devices external to the information processing device 20. The communication device 23 may perform wired communication or wireless communication.

[0030] The input device 24 is a device capable of accepting information input to the information processing device 20 from outside. The input device 24 may include an operation device (e.g., a keyboard, a mouse, a touch panel, etc.) that can be operated by a user of the information processing device 20. The input device 24 may include a recording medium reading device that can read information recorded on a recording medium that is detachable from the information processing device 20, such as a USB (Universal Serial Bus) memory. Note that when information is input to the information processing device 20 via the communication device 23 (in other words, when the information processing device 20 acquires information via the communication device 23), the communication device 23 may function as an input device.

[0031] The output device 25 is a device capable of outputting information to the outside of the information processing device 20. The output device 25 may output visual information such as text or images, auditory information such as sound, or tactile information such as vibration, as the information. The output device 25 may include, for example, at least one of a display, a speaker, a printer, and a vibration motor. The output device 25 may be capable of outputting information to a recording medium detachable from the information processing device 20, such as a USB memory. Note that when the information processing device 20 outputs information via the communication device 23, the communication device 23 may function as the output device.

[0032] 3 shows an example of logical functional blocks realized in the arithmetic device 21 to execute the information processing method. As shown in FIG. 3, a compression unit 211, a creation unit 212, a prediction unit 213, and a selection unit 214 are realized in the arithmetic device 21. The "compression unit 211" is a component corresponding to the "compression unit 11" in the first embodiment described above, the "creation unit 212" is a component corresponding to the "creation unit 12" in the first embodiment described above, and the "prediction unit 213" is a component corresponding to the "prediction unit 13" in the first embodiment described above. [2-2: Information Processing Method Executed by Information Processing Device 20]

[0033] The information processing device 20 is configured as a device for acquiring a predictive prompt for predicting an appropriate recommended item. The information processing device 20 may create a plurality of predictive prompts, perform prediction using each of the plurality of predictive prompts, and select a predictive prompt for predicting an appropriate recommended item based on the prediction result. The predictive prompt includes information indicating an instruction, user information, and test information indicating recommendation candidate items that are candidates for the recommended item to be recommended. The information processing device 20 may determine whether the predictive prompt can predict an appropriate recommended item when it includes which user information and which test information. In other words, the information processing device 20 determines how the user information is compressed and how many recommendation candidate items the test information includes when it can predict an appropriate recommended item.

[0034] The information processing method executed by the information processing device 20 will be described with reference to Fig. 4 to Fig. 6. Fig. 4 is a flowchart showing an example of the flow of the information processing method executed by the information processing device 20. Fig. 5 is a block diagram showing an example of the flow of information in the information processing method executed by the information processing device 20. Fig. 6 is a block diagram illustrating an outline of the operation of the prediction unit 213.

[0035] As shown in Fig. 4, the compression unit 211 generates user information by compressing past performance information indicating the user's past performance (step S21). As shown in Fig. 5, the compression unit 211 may acquire past performance information 222 and generate user information 223. The past performance information 222 may be stored in the storage device 22. The compression unit 211 may also store the generated user information 223 in the storage device 22.

[0036] The compression unit 211 may compress the past performance information by, for example, extracting representative information from the information included in the past performance information. The compression unit 211 may also generate information that is closely related to the information included in the past performance information as user information. A method for compressing the past performance information will be described in another embodiment.

[0037] The compression unit 211 may generate one piece of user information for one user. Alternatively, the compression unit 211 may generate multiple pieces of user information for one user. For example, the compression unit 211 may generate first user information U1_1 and second user information U1_2 for user U1.

[0038] The operation of step S21 may be performed periodically. For example, the compression unit 211 may generate user information by compressing past performance information for the past week every week, and store the generated user information in the storage device 22.

[0039] 4, the user information acquisition unit 2121 acquires user information (step S22). As shown in FIG. 5, the user information acquisition unit 2121 may acquire user information 223 stored in the storage device 22.

[0040] As shown in Fig. 4, the test information acquisition unit 2122 acquires test information indicating recommendation candidate items that are candidates for recommended items (step S23). As shown in Fig. 5, the test information acquisition unit 2122 may acquire test information from the recommendation candidate items 224 stored in the storage device 22. The test information acquisition unit 2122 may acquire test information including a predetermined number of recommendation candidate items. The test information acquisition unit 2122 may change the number of recommendation candidate items included in the test information each time it acquires test information. The number of recommendation candidate items included in the test information is the number of recommendation candidate items to be processed collectively and corresponds to the batch size.

[0041] 4, the creation unit 212 creates a predictive prompt including user information and test information (step S24). If the test information includes a predetermined number of recommendation candidate items, the creation unit 212 may create a predictive prompt for collectively predicting the predetermined number of recommendation candidate items.

[0042] The creation unit 212 may generate a plurality of predictive prompts including the user information and the test information. In this case, the creation unit 212 may generate each of the plurality of predictive prompts such that at least one of the user information and the test information included in the predictive prompt is different from the other.

[0043] As illustrated in FIG. 6 , a predictive prompt may include an instruction, user information, and test information indicating candidate items for recommendation. The instruction included in each of the multiple predictive prompts may be a common instruction. For example, a first predictive prompt may include first user information U1_1 and test information indicating five candidate items for recommendation. For example, a second predictive prompt may include first user information U1_1 and test information indicating ten candidate items for recommendation. For example, a third predictive prompt may include second user information U1_2 and test information indicating five candidate items for recommendation.

[0044] As shown in Fig. 5, the prediction unit 213 acquires the predictive prompt created by the creation unit 212. As shown in Figs. 4 and 5, the prediction unit 213 predicts a recommended item using the LLM using the predictive prompt created by the creation unit 212 (step S25). The prediction unit 213 may predict whether the user likes a recommendation candidate item, and if the user likes a recommendation candidate item, may predict that the item is a recommended item. Note that the LLM may be pre-trained, and may not be trained for information processing according to this embodiment.

[0045] As illustrated in Fig. 6 , the first prediction prompt includes recommendation candidate items with item IDs 100 to 104 in the test information. The LLM that has received the first prediction prompt may output information indicating whether the recommendation candidate items with item IDs 100 to 104 are recommended items. For example, the LLM may output information in which a flag "1" is associated with the recommendation candidate item if it is a recommended item, and a flag "0" is associated with the recommendation candidate item if it is not a recommended item. Fig. 6 illustrates a case in which the LLM that has received the first prediction prompt outputs a prediction result that the recommendation candidate items with item IDs 100 and 104 are recommended items.

[0046] As shown in Fig. 6, the second prediction prompt includes recommendation candidate items with item IDs 105 to 114 in the test information. The LLM that has received the second prediction prompt may output information indicating whether the recommendation candidate items with item IDs 105 to 114 are recommended items. Fig. 6 illustrates a case where the LLM that has received the second prediction prompt outputs a prediction result that the recommendation candidate items with item IDs 105 to 114 are not recommended items.

[0047] As shown in Fig. 6, the third prediction prompt includes, in the test information, recommendation candidate items with item IDs 295 to 299. The LLM that has received the third prediction prompt may output information indicating whether the recommendation candidate items with item IDs 295 to 299 are recommended items. Fig. 6 illustrates a case where the LLM that has received the third prediction prompt outputs a prediction result that the recommendation candidate item with item ID 296 is a recommended item.

[0048] As shown in Fig. 4, the prediction unit 213 stores the predicted prompt and the prediction result (step S26). As shown in Fig. 5, the prediction unit 213 may store the predicted prompt and the prediction result 225 in the storage device 22.

[0049] 4 , the selection unit 214 determines whether to continue repeating the prompt creation operation of the creation unit 212 and the recommended item prediction operation of the prediction unit 213 (step S27). The selection unit 214 may determine to end the repetition of the creation operation and the prediction operation, for example, when all of the recommendation candidate items 224 stored in the storage device 22 are included in any of the predicted prompts. The selection unit 214 may also determine to end the repetition of the creation operation and the prediction operation, for example, when the number of predicted prompts and prediction results 225 stored in the storage device 22 reaches a predetermined number. If it is determined to continue the repetition of the creation operation and the prediction operation (step S27: Yes), the process may return to step S22.

[0050] If it is determined not to continue repeating the creation operation and the prediction operation (step S27: No), the selection unit 214 selects a prediction prompt (referred to as an "appropriate prompt") appropriate for predicting recommended items from the multiple prediction prompts (step S28). The selection unit 214 may select an appropriate prompt from prediction prompts that combine various user information and various test information. The selection unit 214 may select an appropriate prompt by determining which combination of user information and test information is appropriate for predicting recommended items. In other words, the selection unit 214 determines which compression method should be used to compress the past performance information into user information and how many recommendation candidate items should be included in the test information to enable appropriate prediction. Creating a prediction prompt that includes user information of the compression method selected by the selection unit 214 and test information including recommendation candidate items in the number of recommendation candidate items selected by the selection unit 214 enables appropriate recommendations.

[0051] The selection unit 214 may select an appropriate prompt based on the prediction result by the prediction unit 213. For example, there may be about 1,000 pieces of past performance information for the recommendation candidate items. In this case, the selection unit 214 may select an appropriate prompt based on the prediction result by the prediction unit 213 and the past performance information for the recommendation candidate items. In other words, the about 1,000 pieces of past performance information may be used as learning data.

[0052] For example, to learn recommendations using matrix decomposition or the like, information on about 1,000 past performance items is insufficient, and information on about 100,000 past performance items is required. In contrast, the information processing device 20 according to this embodiment can make appropriate recommendations by using past performance information in an amount that is insufficient for learning recommendations using matrix decomposition or the like. [2-4: Technical Effects]

[0053] There is a demand for predicting whether a new product, etc., that has not yet been released and therefore has not yet been purchased by anyone and has no sales record, will sell. For example, when information on service usage history including user IDs, item IDs, and information indicating reactions is available, a recommendation system is known that learns the correspondence between reactions and combinations of user IDs and item IDs, and predicts how a specific user will react when an item is recommended to them. Such a recommendation system requires the provision of information on service usage history. Furthermore, it is difficult to make predictions for items with no purchase history, such as new products.

[0054] In contrast, by utilizing the LLM, it is possible to predict a user's response to an item, such as a new product, by providing prompts about the user and the item, even for items with no purchase history. In other words, by utilizing the LLM, it is possible to predict a user's response to an item, such as a new product, without special training, even for items with no purchase history. On the other hand, when using the LLM, costs are incurred depending on the token length of the prompt provided.

[0055] In other words, costs can be reduced by reducing the token length of the predictive prompt. The information processing device 20 according to the present disclosure can reduce the token length of the predictive prompt and reduce the cost of prediction by compressing the user's past performance information included in the predictive prompt.

[0056] Furthermore, prediction costs are incurred in proportion to the number of prediction processes. Therefore, the prediction costs can be reduced by reducing the number of prediction processes. The information processing device 20 performs batch processing to predict a predetermined number of recommendation candidate items, thereby reducing the number of prediction processes and the prediction costs. [3: Third Embodiment]

[0057] A third embodiment of an information processing device, an information processing method, and a recording medium will be described with reference to FIGS. 7 to 9. Hereinafter, the third embodiment of an information processing device, an information processing method, and a recording medium will be described using an information processing device 30. The third embodiment differs from the first and second embodiments in the method of compressing past performance information. Note that, for the third embodiment, descriptions that overlap with the descriptions of the first and second embodiments will be omitted as appropriate. Note that, in the drawings, parts common to the first and second embodiments are designated by the same reference numerals.

[0058] 7, the arithmetic unit 21 included in the information processing device 30 includes, as logical functional blocks, a compression unit 311, a creation unit 212, a prediction unit 213, and a selection unit 214. The creation unit 212 may include a user information acquisition unit 2121 and a test information acquisition unit 2122. The compression unit 311 may include a past performance item extraction unit 3111 and a summary prompt creation unit 3112. [3-1: Information Processing Method Executed by the Information Processing Device 30]

[0059] The information processing device 30 may generate user information using the LLM. The information processing device 30 may generate, as user information, a summary that summarizes past performance information using the LLM. An information processing method executed by the information processing device 30 will be described with reference to FIGS. 8 and 9. FIG. 8 is a flowchart showing an example of the flow of the information processing method executed by the information processing device 30. FIG. 9 is a block diagram showing an example of the flow of information in the information processing method executed by the information processing device 30.

[0060] 8 and 9, the past performance item extraction unit 3111 extracts K past performance items from the plurality of past performance items included in the past performance item group 222 (step S31). K is a predetermined number, and may be any integer.

[0061] The past performance item extraction unit 3111 may select the K past performance items using any method. The past performance item extraction unit 3111 may also select the K past performance items randomly. The past performance item extraction unit 3111 may also extract a predetermined number of past performance items based on the attributes of each of the multiple past performance items included in the past performance item group 222. For example, the past performance item extraction unit 3111 may select K past performance items from items in the same category, and select K past performance items for each category. The past performance item extraction unit 3111 may extract information on the K past performance items (referred to as "extracted item information") from the past performance information.

[0062] The past performance item extraction unit 3111 may extract multiple pieces of extracted item information. The past performance item extraction unit 3111 may extract multiple pieces of extracted item information without changing the value of K. The past performance item extraction unit 3111 may extract multiple pieces of extracted item information by changing the value of K. Alternatively, the past performance item extraction unit 3111 may extract all of the multiple past performance items included in the past performance item group 222.

[0063] The summary prompt creation unit 3112 creates a summary prompt that summarizes the information on the K past performance items (step S32). The summary prompt is a prompt for generating a summary sentence as user information. When the past performance item extraction unit 3111 extracts multiple pieces of extracted item information, the summary prompt creation unit 3112 may create a summary prompt corresponding to each piece of extracted item information.

[0064] The compression unit 311 uses the summary prompt to generate a summary sentence summarizing the information on the K past performance items extracted by the LLM as user information (step S33). If the summary prompt generation unit 3112 generates multiple summary prompts, the compression unit 311 may generate multiple summary sentences as user information. The compression unit 311 may store the generated summary sentence in the storage device 22 as user information 223.

[0065] The user information acquisition unit 2121 acquires user information (step S22). The test information acquisition unit 2122 acquires test information indicating candidate items for recommendation (step S23). The creation unit 212 creates a predictive prompt including the user information and the test information (step S24). The prediction unit 213 predicts a recommended item using the predictive prompt created by the creation unit 212 with the LLM (step S25). The prediction unit 213 stores the predictive prompt and the prediction result (step S26). If it is determined not to continue the repetition of the creation operation and the prediction operation (step S27: No), the selection unit 214 selects an appropriate prompt appropriate for predicting a recommended item from the multiple predictive prompts (step S28). In other words, the selection unit 214 determines which summary sentence summarizing information on K past performance items is appropriate as user information. [3-2: Technical Effects]

[0066] The information processing device 30 according to this disclosure generates a summary of the user's past history and compresses the past performance information, thereby shortening the predictive prompt. This reduces the processing load. [4: Fourth Embodiment]

[0067] A fourth embodiment of an information processing device, an information processing method, and a recording medium will be described with reference to FIGS. 10 to 12. Hereinafter, the fourth embodiment of an information processing device, an information processing method, and a recording medium will be described using an information processing device 40. The fourth embodiment differs from the first to third embodiments in the method of compressing past performance information. Note that, for the fourth embodiment, descriptions that overlap with the descriptions of the first to third embodiments will be omitted as appropriate. Note that, in the drawings, parts common to the first to third embodiments are designated by the same reference numerals.

[0068] 10, the arithmetic device 21 included in the information processing device 40 includes, as logical functional blocks, a compression unit 411, a creation unit 212, a prediction unit 213, and a selection unit 214. The creation unit 212 may include a user information acquisition unit 2121 and a test information acquisition unit 2122. The compression unit 411 may include a past performance item extraction unit 4111 and a similarity calculation unit 4113. [4-1: Information Processing Method Executed by the Information Processing Device 40]

[0069] The information processing device 40 may generate user information including past performance items that contribute to the recommendation of an item. The information processing device 40 may generate user information by extracting past performance items that contribute to the recommendation of an item. The information processing device 40 may generate user information by extracting past performance items similar to recommendation candidate items. An information processing method executed by the information processing device 40 will be described with reference to FIGS. 11 and 12. FIG. 11 is a flowchart showing an example of the flow of the information processing method executed by the information processing device 40. FIG. 12 is a block diagram showing an example of the flow of information in the information processing method executed by the information processing device 40.

[0070] 11 and 12 , the similarity calculation unit 4113 calculates the similarity between each of the multiple past performance items included in the past performance item group 222 and each of the multiple recommendation candidate items included in the recommendation candidate item group 224 (step S41). First, the past performance item extraction unit 4111 may convert each of the past performance items included in the past performance item group 222 and each of the recommendation candidate items 224 included in the recommendation candidate item group into text and vectorize the text. The similarity calculation unit 4113 may store each of the vectorized past performance items (referred to as a "past performance vector") and each of the vectorized recommendation candidate items (referred to as a "recommendation candidate vector") in the storage device 22. Next, the similarity calculation unit 4113 may calculate the similarity between each of the past performance vectors and each of the recommendation candidate vectors. For example, the similarity calculation unit 4113 may calculate the cosine similarity between each of the past performance vectors and each of the recommendation candidate vectors.

[0071] The past performance item extraction unit 4111 extracts K past performance items from the multiple past performance items based on each similarity (step S42). K is a predetermined number and may be any integer. The past performance item extraction unit 4111 may extract the top K past performance items with the highest similarity. The past performance item extraction unit 4111 may extract the top K past performance items with the highest maximum similarity between the past performance vector and each of the recommendation candidate vectors. In other words, the past performance item extraction unit 4111 extracts past performance items related to the recommendation candidate items.

[0072] The compression unit 411 generates user information including information indicating the K past performance items (step S43). Alternatively, the compression unit 411 may summarize the extracted K past performance items using the summarization method of the third embodiment and further compress the summarization to generate user information.

[0073] The user information acquiring unit 2121 acquires user information (step S22). The test information acquiring unit 2122 acquires test information indicating recommendation candidate items that are candidates for recommended items (step S23). The creating unit 212 creates a predictive prompt including the user information and the test information (step S24). The creating unit 212 in the fourth embodiment may include past performance items related to the recommendation candidate items in the predictive prompt.

[0074] The prediction unit 213 predicts a recommended item using the predictive prompt created by the creation unit 212 through the LLM (step S25). The prediction unit 213 stores the predicted prompt and the prediction result (step S26). If it is determined not to continue repeating the creation operation and the prediction operation (step S27: No), the selection unit 214 selects an appropriate prompt that is appropriate for predicting a recommended item from the plurality of predictive prompts (step S28).

[0075] The extraction of past achievement items in the third embodiment may employ the extraction method of past achievement items in this embodiment. [4-2: Technical Effects]

[0076] The information processing device 40 according to this disclosure extracts past performance items similar to recommendation candidate items and generates user information, so that the user information includes past performance items that contribute to the recommendation of an item, and it is possible to create predictive prompts that can make appropriate recommendations. [5: Fifth Embodiment]

[0077] A fifth embodiment of an information processing device, an information processing method, and a recording medium will be described with reference to FIGS. 13 to 15 . Hereinafter, the fifth embodiment of an information processing device, an information processing method, and a recording medium will be described using an information processing device 50. The fifth embodiment differs from the first to fourth embodiments in the method of selecting an appropriate prompt. Note that, for the fifth embodiment, descriptions that overlap with the descriptions of the first to fourth embodiments will be omitted as appropriate. Note that, in the drawings, parts common to the first to fourth embodiments are designated by the same reference numerals.

[0078] 13, the arithmetic device 21 included in the information processing device 50 includes, as logical functional blocks, a compression unit 211, a creation unit 212, a prediction unit 513, and a selection unit 514. The creation unit 212 may include a user information acquisition unit 2121 and a test information acquisition unit 2122. The selection unit 514 may include a cost acquisition unit 5141 and a score acquisition unit 5142. [5-1: Information Processing Method Executed by the Information Processing Device 50]

[0079] The information processing device 50 may be configured as a device that selects an appropriate predictive prompt based on the predicted cost when the corresponding predictive prompt is used and the score of the prediction result when the corresponding predictive prompt is used. An information processing method executed by the information processing device 50 will be described with reference to FIGS. 14 to 17 . FIG. 14 is a block diagram showing an example of the flow of information in the information processing method executed by the information processing device 50. FIG. 15 is a block diagram showing an example of score calculation. FIGS. 16 and 17 are tables for explaining the selection of an appropriate prompt based on the score and cost.

[0080] As shown in FIG. 14 , the prediction unit 513 may output the prediction cost along with the prediction result. The prediction cost may be a value corresponding to the time spent on prediction using the prediction prompt. The prediction cost may also be a value corresponding to the usage cost of an API (Application Programming Interface) or the token length of the prediction prompt. The usage cost of the API is approximately equal to the token length of the prediction prompt. Therefore, the prediction cost may be a value corresponding to the usage cost of the API. In other words, the prediction cost may be a value corresponding to at least one of the time spent on prediction using the prediction prompt and the token length of the prediction prompt. The prediction unit 513 may store the prediction prompt, the prediction result, and the prediction cost 525 in the storage device 22.

[0081] The cost acquisition unit 5141 acquires the predicted cost C(U, n) of the predicted prompt P(U, n) including user information U and test information including information on n recommendation candidate items. The cost acquisition unit 5141 may acquire the predicted cost C(U, n) stored in the storage device 22.

[0082] The score acquiring unit 5142 acquires a score S(U, n) representing the performance of a prediction prompt P(U, n) including user information U and test information including information on n recommendation candidate items. The score acquiring unit 5142 may acquire the score S(U, n) calculated from precision, recall, F1 score, AUC (Area Under the ROC Curve), nDCG (Normalized Documented Cumulative Gain), etc.

[0083] The score S(U, n) may be calculated based on the predicted results of items to be recommended to the user and the recommendation results when items are recommended to the user. In the fifth embodiment, information on past performance of the recommendation candidate items may exist. For example, information on past performance of approximately 1,000 recommendation candidate items may exist. In other words, information on past performance of approximately 1,000 items may be used as learning data.

[0084] 15 , for example, the first prediction result and past performance match for Item IDs "100," "102," and "103," but do not match for Item ID "101." In this case, the score S(U, n) may be calculated to be lower than when the first prediction result and past performance match for Item IDs "100," "102," and "103," as well as for Item ID "101." Furthermore, the score S(U, n) may be calculated to be higher than when the first prediction result and past performance do not match for any of Item IDs "100," "102," and "103."

[0085] Performance and cost are important considerations when predicting recommended items. As mentioned above, compressing past performance information or increasing the batch size can reduce costs. On the other hand, excessive compression can have a negative impact on prediction performance. Increasing the number of items in a batch process often reduces total cost but reduces performance. In other words, prioritizing cost reduces performance. Conversely, a long prediction prompt often improves performance but increases cost. In other words, prioritizing performance increases cost. Therefore, it is necessary to appropriately select how to compress the user's past performance information and what the batch size should be. The selection unit 514 selects a prediction prompt P(U, n) to reduce costs while maintaining a certain level of performance. The selection method adopted by the selection unit 514 is arbitrary. Two example selection methods are described below. [5-2: Example Method 1]

[0086] In the example method 1, the selector 514 introduces a threshold p, which can take a value between 0 and 1. The selector 514 determines a lower limit for the score S by multiplying the maximum value of each score S(U,n) corresponding to each predicted prompt P(U,n) by the threshold p. The selector 514 selects the predicted prompt P(U,n) with the smallest cost C(U,n) from the predicted prompts P(U,n) with scores S(U,n) greater than the lower limit for the score S.

[0087] A specific example of the case where the example method 1 is adopted will be described with reference to FIG. 16 . FIG. 16 illustrates an example in which the score S(U1,5) when using the predicted prompt (U1,5) is "0.8" and the cost C(U1,5) when using the predicted prompt (U1,5) is "100." The predicted prompt (U1,5) includes user information U1 and test information with a batch size of 5 (the number of recommendation candidate items is 5). FIG. 16 also illustrates an example in which the score S(U1,10) when using the predicted prompt (U1,10) is "0.72" and the cost C(U1,10) when using the predicted prompt (U1,10) is "50." The predicted prompt (U1,10) includes user information U1 and test information with a batch size of 10 (the number of recommendation candidate items is 10). FIG. 16 also illustrates an example in which the score S(U1,20) when using the predicted prompt (U1,20) is "0.71" and the cost C(U1,20) when using the predicted prompt (U1,20) is "30." The predicted prompt (U1,20) includes user information U1 and test information with a batch size of 20 (the number of recommendation candidate items is 20). FIG. 16 also illustrates an example in which the score S(U5,10) when using the predicted prompt (U5,10) is "0.78" and the cost C(U5,10) when using the predicted prompt (U5,10) is "60." The predicted prompt (U5,10) includes user information U5 and test information with a batch size of 10.

[0088] 16 , the selector 514 multiplies the score S(U1,5) (=0.8) corresponding to the predicted prompt (U1,5) with the highest score S among the predicted prompts P(U,n) by the threshold p (=0.95) to calculate 0.76. Of the predicted prompts P(U,n) with scores S(U,n) greater than 0.76, the selector 514 selects the predicted prompt P(U,n) with the smallest cost C(U,n). In the example shown in FIG. 16, the selector 514 selects, as an appropriate prompt, the predicted prompt P(U5,10) (cost C(U5,10)=60) with the smallest cost C(U,n) from the predicted prompts P(U1,5) (score S(U1,5)=0.8) and P(U5,10) (score S(U5,10)=0.78) with scores S(U,n) greater than 0.76. [5-3: Example Method 2]

[0089] In Example Method 2, the selection unit 514 selects an appropriate prompt using the ranking of each score S(U,n) of each predicted prompt (U,n) (referred to as the “score ranking”) and the ranking of each cost C(U,n) of each predicted prompt (U,n) (referred to as the “cost ranking”).

[0090] 17 illustrates a case in which the score ranking of predicted prompt P(U1,5) is 1st, the score ranking of predicted prompt P(U1,10) is 7th, the score ranking of predicted prompt P(U1,20) is 8th, and the score ranking of predicted prompt P(U5,10) is 2nd. Also, FIG. 17 illustrates a case in which the cost ranking of predicted prompt P(U1,5) is 20th, the cost ranking of predicted prompt P(U1,10) is 5th, the cost ranking of predicted prompt P(U1,20) is 4th, and the cost ranking of predicted prompt P(U5,10) is 8th.

[0091] The selector 514 may calculate an overall ranking from the score ranking and the cost ranking and select the predicted prompt P(U,n) with the highest overall ranking. For example, the selector 514 may calculate the overall ranking by adding the score ranking and the cost ranking. The “Rank Total” column in FIG. 17 illustrates a case where the predicted prompt P(U1,5) has a rank total of “21,” the predicted prompt P(U1,10) has a rank total of “12,” the predicted prompt P(U1,20) has a rank total of “12,” and the predicted prompt P(U5,10) has a rank total of “10.” In this case, the selector 514 may select the predicted prompt P(U5,10) with the smallest rank total (rank total = 10) as the appropriate prompt.

[0092] Alternatively, the selector 514 may weight the score ranking and the cost ranking and add them together to determine an overall ranking. FIG. 17 illustrates an example in which the selector 514 calculates a weighted ranking total by adding the score ranking and twice the cost ranking. The “Weighted Rank Total” column in FIG. 17 illustrates an example in which the weighted ranking total of the predicted prompt P(U1,5) is “41,” the weighted ranking total of the predicted prompt P(U1,10) is “17,” the weighted ranking total of the predicted prompt P(U1,20) is “16,” and the weighted ranking total of the predicted prompt P(U5,10) is “18.” In this case, the selector 514 may select the predicted prompt P(U1,20) with the smallest weighted ranking total (weighted ranking total = 16) as the appropriate prompt.

[0093] The selection unit 514 may also calculate a mean reciprocal rank (MRR) as an overall ranking, which is an average of the reciprocal of the score ranking and the reciprocal of the cost ranking. [5-4: Technical Effects] The information processing device 50 according to this embodiment can select predictive prompts that can reduce costs without degrading performance. [6: Sixth Embodiment]

[0094] A sixth embodiment of an information processing device, an information processing method, and a recording medium will be described with reference to FIGS. 18 and 19 . Hereinafter, the sixth embodiment of an information processing device, an information processing method, and a recording medium will be described using an information processing device 60. The sixth embodiment differs from the first to fifth embodiments in that a prompt corresponding to a selected appropriate prompt is created and a recommended item is predicted. Note that, for the sixth embodiment, descriptions that overlap with the descriptions of the first to fifth embodiments will be omitted as appropriate. Note that, in the drawings, parts common to the first to fifth embodiments are denoted by the same reference numerals.

[0095] 18, the arithmetic unit 21 included in the information processing device 60 includes, as logical functional blocks, a compression unit 611, a creation unit 612, and a prediction unit 613. The creation unit 612 may include a test information acquisition unit 6122. [6-1: Information Processing Method Executed by the Information Processing Device 60]

[0096] The information processing device 60 may be configured as a device that creates a predictive prompt based on user information and test information included in an appropriate prompt selected by at least one of the information processing devices 20 according to the second embodiment to the information processing device 50 according to the fifth embodiment, and recommends recommended items to the user. In other words, the information processing device 60 may be configured as a device that distributes the selection result by at least one of the information processing devices 20 according to the second embodiment to the information processing device 50 according to the fifth embodiment to a user who uses a service. In this embodiment, the selected combination of compression method and batch size is distributed to the user who uses the service, and an item appropriate for the user who uses the service is recommended. The information processing device 60 may also collect user information by compressing past performance information using an appropriate compression method.

[0097] An information processing method executed by the information processing device 60 will be described with reference to Fig. 19. Fig. 19 is a flowchart showing an example of the flow of the information processing method executed by the information processing device 60.

[0098] 19 , if there is a user (referred to as a "target user") who uses the service and to whom an item should be recommended (step S61: Yes), the compression unit 611 generates user information (referred to as "target user information") by compressing the past performance information of the target user (step S62). In this embodiment, the compression unit 611 may compress the past performance information of the target user using a compression method for user information included in the appropriate prompt selected by at least one of the information processing device 20 according to the second embodiment to the information processing device 50 according to the fifth embodiment, and generate the target user information.

[0099] The test information acquisition unit 6122 acquires test information (referred to as "target test information") (step S63). The test information acquisition unit 6122 may acquire recommendation candidate items using a method for acquiring recommendation candidate items included in test information included in an appropriate prompt selected by at least one of the information processing device 20 according to the second embodiment to the information processing device 50 according to the fifth embodiment. The test information acquisition unit 6122 may acquire target test information of a batch size of the test information included in the appropriate prompt selected by at least one of the information processing device 20 according to the second embodiment to the information processing device 50 according to the fifth embodiment. The creation unit 612 creates a predicted prompt (referred to as a "target predicted prompt") including target user information and the target test information (step S64).

[0100] The prediction unit 613 uses the target prediction prompt to predict a recommended item (referred to as a "target recommended item") to be recommended to the target user in the LLM, and outputs the target recommended item (step S65). [6-2: Technical Effects]

[0101] The information processing device 60 creates a prediction prompt including the user's past performance information compressed by an appropriate compression method and an appropriate batch size of recommendation candidate items, and uses these for prediction, so that it is possible to predict recommended items with high accuracy and at low cost. [Example 1: Operation of an Electronic Commerce (EC) Site]

[0102] When an item is a new product that has just been released and there are not enough logs related to the item, it may be necessary to determine to whom the item should be recommended. An EC site can accumulate the behavioral history (past performance) of users who use the EC site. When a user logs in, the EC site using this embodiment can make highly accurate inferences about new products based on the user's past performance at no cost.

[0103] The recommended items in this embodiment may be products such as daily necessities, food, books, and home appliances. The recommended items in this embodiment may also be stores. For example, in a credit card company, the information processing device 60 may be applied to predict which store a user should visit. [Specific Example 2: Planning a Policy for a Company]

[0104] As mentioned above, a user may be an individual or a corporation. A corporation may be a company. In this case, the recommended items may be measures proposed to the company. A consultant or the like may have the opportunity to propose measures to a client company. The measures may not yet have a past track record. For example, the client retail store and other companies in the same industry as the client retail store have never held the desired exhibition or sale event. The client retail store has held multiple exhibitions and sales events and has records of their results. The records of results may be obtained from the client retail store. The records of results may also be obtained from past articles about the client retail store, for example, using the Internet.

[0105] The information processing device 60 generates compressed information on the past performance information of the client retail store and, based on the compressed information, can predict whether or not a new measure should be proposed. Since there are no logs like in e-commerce, the construction of a general learning-based recommendation system is not suitable for proposing measures. The measures are customized for the client company. In other words, the measures are cold-start items with no logs. Therefore, they are suitable for LLM recommendations.

[0106] The number of measures increases with the combination of elements. For example, a huge number of proposed measures can be created, such as {for people in their 30s, for people in their 40s, ...} x {for men, for women} x {beauty-conscious, health-conscious, ...} x .... For this reason, efficient recommendations using LLM are desired. By applying the information processing device 60 according to this embodiment, it is possible to make LLM recommendations more efficient. [7: Supplementary Note]

[0107] Some or all of the above embodiments may be described as, but are not limited to, the following notes. [Note 1] An information processing device comprising: a compression unit that generates user information obtained by compressing past performance information indicating a user's past performance; a creation unit that creates a predictive prompt including the user information and for predicting recommended items to be recommended to the user; and a prediction unit that uses the predictive prompt to predict the recommended items using a large language model (LLM) and output the recommended items. [Note 2] The creation unit creates a plurality of predictive prompts including the user information and test information indicating candidate recommendation items that are candidates for the recommended items, and each of the plurality of predictive prompts differs in at least one of the user information and the test information included in the predictive prompt, and the information processing device comprises a selection unit that selects an appropriate prompt appropriate for predicting the recommended items from the plurality of predictive prompts. [Note 3] The information processing device according to Note 1, wherein the compression unit uses a large language model (LLM) to generate a summary that summarizes the past performance information as the user information. [Supplementary Note 4] The information processing device according to Supplementary Note 1, wherein the compression means comprises: a calculation means for calculating a similarity between each of a plurality of past performance items included in the past performance information and each of a plurality of recommendation candidate items that are candidates for the recommended item; and an extraction means for extracting a predetermined number of past performance items from the plurality of past performance items based on each of the similarities. [Supplementary Note 5] The information processing device according to Supplementary Note 3, wherein the compression means extracts past performance items included in the past performance information, creates a summary prompt including the extracted past performance items, and generates a summary sentence summarizing the past performance items as the user information using a Large Language Model (LLM) by using the summary prompt.[Supplementary Note 6] The information processing device according to Supplementary Note 4, wherein the calculation means calculates the similarity between each of a plurality of past performance vectors obtained by converting each of the plurality of past performance items into text and vectorizing it, and each of a plurality of recommendation candidate vectors obtained by converting each of the plurality of recommendation candidate items into text and vectorizing it. [Supplementary Note 7] The information processing device according to Supplementary Note 2, wherein the selection means selects the appropriate prompt based on a prediction result by the prediction means. [Supplementary Note 8] The information processing device according to Supplementary Note 2, wherein the selection means selects the appropriate prompt based on a prediction cost when the prediction prompt is used and a score of prediction accuracy when the prediction prompt is used. [Supplementary Note 9] The information processing device according to Supplementary Note 5, wherein the compression means extracts all of a plurality of past performance items included in the past performance information. [Supplementary Note 10] The information processing device according to Supplementary Note 5, wherein the compression means extracts a predetermined number of past performance items from a plurality of past performance items included in the past performance information. [Supplementary Note 11] The information processing device according to Supplementary Note 5, wherein the compression means extracts a predetermined number of past performance items from the plurality of past performance items based on attributes of each of the plurality of past performance items included in the past performance information. [Supplementary Note 12] The information processing device according to Supplementary Note 8, wherein the cost is calculated based on a processing time required for prediction. [Supplementary Note 13] The information processing device according to Supplementary Note 8, wherein the cost is calculated based on a token length of the prompt. [Supplementary Note 14] The information processing device according to Supplementary Note 8, wherein the score is calculated based on a prediction result of an item to be recommended to the user and a recommendation result when an item is recommended to the user.[Supplementary Note 15] The information processing device according to Supplementary Note 2, wherein the compression means generates target user information by compressing the past performance information of a target user who uses a service using a compression method for the user information included in the appropriate prompt, and the device is equipped with test information acquisition means for acquiring target test information indicating recommendation candidate items that are candidates for recommendation to the target user using a method for acquiring the test information included in the appropriate prompt, the creation means creates a target predicted prompt including the target user information and the target test information, and the prediction means uses the target predicted prompt to predict target recommended items to be recommended to the target user using the LLM, and outputs the target recommended items. [Supplementary Note 16] An information processing method executed by a computer, comprising: generating user information by compressing past performance information indicating a user's past performance, creating a predictive prompt including the user information for predicting recommended items to be recommended to the user, and predicting the recommended items using the predictive prompt using a Large Language Model (LLM), and outputting the recommended items. [Supplementary Note 17] A recording medium having recorded thereon a computer program for causing a computer to execute an information processing method, the method including: generating user information by compressing past performance information indicating the past performance of a user; creating a predictive prompt including the user information and for predicting recommended items to be recommended to the user; and predicting the recommended items using a large language model (LLM) by using the predictive prompt, and outputting the recommended items.

[0108] Furthermore, some or all of the configurations described in Supplementary Notes 2 to 15, which are dependent on Supplementary Notes 1, and Supplementary Notes 16 and 17, may be dependent in the same manner as Supplementary Notes 2 to 15. Furthermore, not limited to Supplementary Notes 1, 16, and 17, some or all of the configurations described as Supplements may be dependent on various hardware, software, various recording means for recording software, or systems, within the scope of each of the above-mentioned embodiments.

[0109] This disclosure may be modified as appropriate within the scope that does not contradict the gist or idea of ​​the invention that can be read from the claims and the entire specification, and information processing devices, information processing methods, and programs that involve such modifications are also included in the technical idea of ​​this disclosure.

[0110] 10, 20, 30, 40, 50, 60 Information processing device 11, 211, 311, 411, 611 Compression unit 12, 212, 612 Creation unit 13, 213, 513, 613 Prediction unit 2121 User information acquisition unit 2122, 6122 Test information acquisition unit 214, 514 Selection unit 3111, 4111 Past performance item extraction unit 3112 Summary prompt creation unit 4113 Similarity calculation unit 5141 Cost acquisition unit 5142 Score acquisition unit

Claims

1. An information processing device comprising: a compression means for generating user information by compressing past performance information indicating a user's past performance; a creation means for creating a predictive prompt including the user information and for predicting recommended items to be recommended to the user; and a prediction means for predicting the recommended items using a large language model (LLM) using the predictive prompt and outputting the recommended items.

2. The information processing device according to claim 1, further comprising a selection means for selecting an appropriate prompt suitable for predicting the recommended item from the plurality of predictive prompts, wherein the creation means creates a plurality of predictive prompts each including the user information and test information indicating candidate items for recommendation that are candidates for the recommended item, and wherein each of the plurality of predictive prompts is different in at least one of the user information and the test information included in the predictive prompt.

3. The information processing apparatus according to claim 1, wherein the compression means uses Large Language Models (LLM) to generate a summary of the past performance information as the user information.

4. The information processing device according to claim 1, wherein the compression means includes: a calculation means for calculating the degree of similarity between each of a plurality of past performance items included in the past performance information and each of a plurality of recommendation candidate items that are candidates for the recommended item; and an extraction means for extracting a predetermined number of past performance items from the plurality of past performance items based on each of the degrees of similarity.

5. The information processing device according to claim 3, wherein the compression means extracts past performance items included in the past performance information, creates a summary prompt including the extracted past performance items, and uses the summary prompt to generate a summary sentence summarizing the past performance items as the user information using a Large Language Model (LLM).

6. The information processing device according to claim 4, wherein the calculation means calculates the similarity between each of a plurality of past performance vectors obtained by converting each of the plurality of past performance items into text and vectorizing it, and each of a plurality of recommendation candidate vectors obtained by converting each of the plurality of recommendation candidate items into text and vectorizing it.

7. The information processing apparatus according to claim 2, wherein the selection means selects the appropriate prompt based on the prediction result by the prediction means.

8. The information processing device according to claim 2, wherein the selection means selects the appropriate prompt based on a prediction cost when the predictive prompt is used and a score of prediction accuracy when the predictive prompt is used.

9. The information processing apparatus according to claim 5, wherein said compression means extracts all of the plurality of past performance items included in said past performance information.

10. An information processing apparatus according to claim 5, wherein said compression means extracts a predetermined number of past performance items from among a plurality of past performance items included in said past performance information.

11. An information processing device according to claim 5, wherein said compression means extracts a predetermined number of past performance items from among said plurality of past performance items based on the attributes of each of said plurality of past performance items included in said past performance information.

12. The information processing device according to claim 8, wherein the cost is calculated based on the processing time required for the prediction.

13. The information processing apparatus of claim 8, wherein the cost is calculated based on a token length of the predicted prompt.

14. The information processing device according to claim 8, wherein the score is calculated based on a predicted result of an item to be recommended to the user and a recommendation result when an item is recommended to the user.

15. The information processing device of claim 2, wherein the compression means generates target user information by compressing the past performance information of the target user who uses the service using a compression method for the user information included in the appropriate prompt, and the information processing device is equipped with a test information acquisition means for acquiring target test information indicating candidate recommendation items that are candidates for recommendation to the target user using a method for acquiring the test information included in the appropriate prompt, the creation means creates a target prediction prompt including the target user information and the target test information, and the prediction means uses the target prediction prompt to predict target recommended items to be recommended to the target user in the LLM and output the target recommended items.

16. A computer-implemented information processing method comprising: generating user information by compressing past performance information indicating a user's past performance; creating a predictive prompt including the user information and for predicting recommended items to be recommended to the user; and using the predictive prompt to predict the recommended items using Large Language Models (LLMs), and outputting the recommended items.

17. A recording medium having recorded thereon a computer program for causing a computer to execute an information processing method, the method comprising: generating user information by compressing past performance information indicating a user's past performance; creating a predictive prompt that includes the user information and predicts recommended items to be recommended to the user; and using the predictive prompt to predict the recommended items using a Large Language Model (LLM) and output the recommended items.

Images