Apparatus and method

Abstract

To obtain undiscovered information for ranking prediction. [Solution] The sales ranking acquisition unit 102 acquires the sales ranking of products. The learning prediction ranking generation unit 103 generates a learning prediction ranking based on the selection tendencies (two-choice predictions) of multiple products generated by the generation AI 200 using prompt generation information. The accuracy rate calculation unit 104 then calculates the accuracy rate by comparing the learning prediction ranking with the sales ranking. The improvement processing unit 105 functions as a prompt information acquisition unit and, based on the sales ranking, learning prediction ranking, and accuracy rate, has the generation AI 200 generate and acquire previously undiscovered information other than taste (preference) used in generating the learning prediction ranking of products, as well as an improved taste profile, and stores them in the prompt generation information storage unit 101.

Description

Technical Field

[0001] The present invention relates to an apparatus and method for predicting the ranking of products.

Background Art

[0002] Patent Document 1 describes ranking products and users according to the popularity of current products, predicting the sales status of products from the purchases and accesses of users by customers, and predicting the popularity trend of new products based on the purchase history and access history of new products.

Prior Art Documents

Patent Documents

[0003]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0004] However, in Patent Document 1, it only predicts the popularity trend from the purchase history and access history, and it is difficult to accurately predict the popularity trend. For example, it is impossible to predict the popularity trend based on unpublished information not disclosed for that product.

[0005] Therefore, an object of the present disclosure is to provide an apparatus and method capable of acquiring unpublished information for ranking prediction.

Means for Solving the Problems

[0006] The apparatus of this disclosure includes: a storage unit that stores prompt generation information including product names and product information of a plurality of products to be predicted, as well as undiscovered information other than preferences used to generate a prediction ranking of the products, and preference profiles used to generate a prediction ranking of the products; a correct ranking acquisition unit that acquires the correct ranking of the products; a learning prediction ranking generation unit that generates a learning prediction ranking using the prompt generation information; an accuracy rate calculation unit that calculates the accuracy rate of the learning prediction ranking by comparing it with the correct ranking; and a prompt information acquisition unit that acquires undiscovered information other than preferences used to generate a learning prediction ranking of the products, and an improved taste profile, based on the correct ranking, the learning prediction ranking, and the accuracy rate, and stores them in the storage unit. [Effects of the Invention]

[0007] According to the present invention, it is possible to obtain unexplored information for ranking prediction. [Brief explanation of the drawing]

[0008] [Figure 1] Figure 1 is a diagram showing the system configuration of the prediction system 100 of this disclosure. [Figure 2] Figure 2 is a block diagram showing the functional configuration of the prediction system 100 of this disclosure. [Figure 3] Figure 3 shows a specific example of product information. [Figure 4] Figure 4 shows a specific example of a taste profile. [Figure 5] Figure 5 shows a specific example of uncharted information. [Figure 6] Figure 6 is a flowchart showing the learning phase processing of the prediction system 100 of this disclosure. [Figure 7] Figure 7 shows a specific example of the improvement history DB300. [Figure 8] Figure 8 shows a specific example of the improvement history DB300. [Figure 9]FIG. 9 is a diagram showing a specific example of a prompt for binary prediction. [Figure 10] FIG. 10 is a schematic diagram showing a binary selection process. [Figure 11] FIG. 11 is a diagram showing the correct answer rate in comparison with other products for each product. [Figure 12] FIG. 12 is a diagram showing a method for calculating a correct answer rate table. [Figure 13] FIG. 13 is a diagram showing an example of aggregating 50 prediction results and a correct answer rate table for each pair. [Figure 14] FIG. 14 is a diagram showing the predicted sales ranking and the correct sales ranking. [Figure 15] FIG. 15 is a diagram showing a binary prediction prompt including an improved taste profile and unopened information in addition to the taste profile. [Figure 16] FIG. 16 is a diagram showing a specific example of an improvement analysis prompt. [Figure 17] FIG. 17 is a diagram showing the functional configuration of the prediction system 100 for executing the test phase process of the present disclosure. [Figure 18] FIG. 18 is a diagram showing a prompt for generating a prediction confidence instruction (including outlier removal). [Figure 19] FIG. 19 is a diagram showing an output result indicating the prediction confidence result. [Figure 20] FIG. 20 is a diagram showing an outlier product output result. [Figure 21] FIG. 21 is a flowchart showing the operation of the prediction system 100 for executing the test phase process. [Figure 22] FIG. 22 is a diagram showing the details of the correct answer rate table with and without outlier removal. [Figure 23] FIG. 23 is a table of the correct answer rate of binary prediction. [Figure 24] FIG. 24 is a block diagram showing the functional configuration of the prediction system 100 in the operation phase. [Figure 25] FIG. 25 is a flowchart showing the operation of the prediction system in the operation phase process. [Figure 26] FIG. 26 is a diagram showing a prediction ranking that does not consider the outlier and a prediction ranking that considers the outlier. [Figure 27] FIG. 27 is a diagram showing an example of the hardware configuration of the prediction system 100 according to an embodiment of the present disclosure. MODE FOR CARRYING OUT THE INVENTION

[0009] Embodiments of the present disclosure will be described with reference to the accompanying drawings. Where possible, the same parts are denoted by the same reference numerals, and redundant descriptions are omitted.

[0010] FIG. 1 is a diagram showing the system configuration of the prediction system 100 of the present disclosure. As shown in the figure, the prediction system 100 includes a processing unit 100a and a prompt generation information storage unit 101. The processing unit 100a operates by being divided into a learning phase process, a test phase process, and an operation phase process. Further, the prompt generation information storage unit 101 stores product information, product names, unopened information, and taste profiles. Details of these information will be described later.

[0011] The prediction system 100 is communicatively connected to a generation AI 200, an improvement history DB 300, and a POS system 400. The prediction system 100 generates and transmits a prompt for a generation instruction to the generation AI 200 and acquires the answer result thereof. This answer result is an answer for predicting a product, such as a sales ranking of the product, the unopened information on which it is based, and an improved taste profile.

[0012] Further, the prediction system 100 stores the answer history of the generation AI 200 of the unopened information and the improved taste profile in the improvement history DB 300.

[0013] Further, the prediction system 100 acquires sales ranking data 500 from the POS system 400.

[0014] The Taste Questionnaire 600 provides information for generating taste profiles. For example, a taste profile is generated by surveying 20,000 Japanese people.

[0015] Figure 2 is a block diagram showing the functional configuration of the prediction system 100 of this disclosure. This Figure 2 shows the functional configuration in the learning phase of the prediction system 100 and is a detailed view of the processing unit 100a in Figure 1.

[0016] As shown in the figure, the prediction system 100 in the learning phase processing includes a prompt generation information storage unit 101, a sales ranking acquisition unit 102, a learning prediction ranking generation unit 103, an accuracy rate calculation unit 104, and an improvement processing unit 105.

[0017] The prompt generation information storage unit 101 is the part that stores prompt generation information. This prompt generation information storage unit 101 stores, for example, product information, product name, unopened information, and taste profile. The product name is the name of the product used in the sales ranking and forecast ranking. For example, "Nori Bento" (seaweed bento).

[0018] Figure 3 is a diagram illustrating a specific example of product information. In this disclosure, the products handled by the bento shop are used as examples. As shown in the figure, this product information exemplifies information describing a nori bento (seaweed bento). In addition, this disclosure includes product information for other types of bento.

[0019] Figure 4 shows a specific example of a taste profile. As shown in the figure, a taste profile is information that indicates taste tendencies. In this disclosure, the information is obtained from a questionnaire given to 20,000 Japanese people, and therefore represents information that indicates the taste tendencies of Japanese people. This taste profile is information that shows the overall taste tendencies that can be gleaned from the above questionnaire. This text is generated by the generation AI200, but may also be generated by a human.

[0020] Figure 5 shows a specific example of unexplored information. This unexplored information is information other than taste, as shown in the taste profile. This unexplored information, like the taste profile, is information that influences product ranking. This unexplored information is not present in the first training run of the prediction system 100, and is generated by the generation AI 200 through training. This unexplored information includes importance, indicating how important the unexplored information is.

[0021] The sales ranking acquisition unit 102 is the part that acquires the sales ranking, which shows the actual sales of the product to be predicted (bento boxes in this disclosure), obtained from the POS system 400. The sales ranking acquisition unit 102 is the part that stores the acquired sales ranking.

[0022] The learning prediction ranking generation unit 103 is the part that executes an inference process to generate a prediction ranking (learning prediction ranking) of the target products during the learning phase. During the learning phase, it is simply referred to as the prediction ranking. In this disclosure, the learning prediction ranking generation unit 103 takes two products from each product listed in the sales ranking, has the generating AI 200 decide which of the two products it likes (i.e., which one has higher sales), and obtains the answer. The learning prediction ranking generation unit 103 then obtains the answer for all combinations and generates a prediction ranking. The learning prediction ranking generation unit 103 may also calculate the overall accuracy rate (accuracy rate in Figure 10(f)) based on the answers obtained when generating the prediction ranking (Figures 10(a) to 10(d)).

[0023] The accuracy rate calculation unit 104 calculates the accuracy rate of the prediction ranking based on the prediction ranking and the sales ranking. Specifically, the accuracy rate calculation unit 104 extracts two products listed in the prediction ranking and has the generating AI 200 determine which product has higher sales. The accuracy rate calculation unit 104 then calculates the accuracy rate for each product in the prediction ranking and generates an accuracy rate table. This detailed processing will be described later. The accuracy rate in the accuracy rate table is distinguished from the overall accuracy rate (accuracy rate in Figure 10(f)).

[0024] The improvement processing unit 105 analyzes the prediction ranking generated by the learning prediction ranking generation unit 103 and the accuracy rate table generated by the accuracy rate calculation unit 104 to generate improvement content. The improvement processing unit 105 analyzes the improvement content by referring to the improvement history in the improvement history DB 300 and stores the analysis results in the improvement history DB 300.

[0025] The operation of the prediction system 100 configured in this way for learning will now be described. Figure 6 is a flowchart of the learning phase processing of the prediction system 100 of this disclosure. In the prediction system 100, the acquisition unit (not shown) acquires product names and product information from the POS system 400 and stores them in the prompt generation information storage unit 101 (S101). In addition, a taste profile is generated by the generation AI 200 based on a taste questionnaire and stored in the prompt generation information storage unit 101 (S102). These processes S101 and S102 are performed in advance.

[0026] The sales ranking acquisition unit 102 then acquires the sales ranking from the POS system 400 (S103). Based on the sales ranking and prompt generation information, the learning prediction ranking generation unit 103 selects two target products from the sales ranking and performs a two-choice prediction to allow the generating AI 200 to predict its preferred product (S104). The learning prediction ranking generation unit 103 performs two-choice predictions until predictions have been completed for all combinations (S105). The learning prediction ranking generation unit 103 generates a prediction ranking using the two-choice prediction, and the accuracy rate calculation unit 104 calculates the accuracy rate (accuracy rate table) by comparing the prediction ranking with the sales ranking (S106). The accuracy rate calculation unit 104 stores the accuracy rate calculated here in the improvement history DB 300, and calculates and stores the improvement rate and the difference in the improvement rate in the improvement history DB 300. The accuracy rate, improvement rate, and the difference between them calculated in the second and subsequent iterations of the inference process are based on the improvements obtained in the immediately preceding improvement process (S108).

[0027] The learning prediction ranking generation unit 103 determines whether or not the best result is determined by Early Stopping (S107).

[0028] If the best solution is not determined by Early Stopping, the improvement processing unit 105 analyzes the improvement content (S108). Specifically, the improvement processing unit 105 refers to the improvement history in the improvement history DB 300, analyzes the improvement content based on it, and stores the analysis results in the improvement history DB 300 (S108). The analysis results are unexplored information and the improved taste profile.

[0029] The improvement processing unit 105 stores the acquired analysis results (unopened information and taste profile) in the improvement history DB 300, and also stores the unopened information and taste profile in the prompt generation information storage unit 101 to update the prompt generation information storage unit 101 (S109).

[0030] Then, the learning prediction ranking generation unit 103 performs a two-choice prediction again using the updated unexplored information and taste profile (S104). The learning prediction ranking generation unit 103 determines the best option by Early Stopping according to the accuracy rate based on the accuracy rate table, and then determines the optimal unexplored information and taste profile (S110).

[0031] Next, we will explain the detailed operation of each process. Figures 7 and 8 show specific examples of the Improvement History DB300. Figure 7 shows the improved taste profile portion of the Improvement History DB300, and Figure 8 shows information on areas where improvement has not yet been achieved and information on differences from the previous time. For the sake of diagramming, it is divided into two figures, but it could also be a single table associated with each improvement count. As shown in the figures, the Improvement History DB300 stores the number of improvements, the accuracy rate, the improvement rate from the baseline, the improved taste profile, information on areas where improvement has not yet been achieved, the difference in the improvement rate from the previous time, and the difference in the improvement content from the previous time.

[0032] As described above, the improvement history DB300 stores the accuracy rate calculated by the learning prediction ranking generation unit 103 and the accuracy rate calculation unit 104. The number of improvements is the number of times the improvement processing process was executed. The accuracy rate is the percentage of times the prediction result for the sales ranking in the prediction ranking at that time was correct, and is the average value in the accuracy rate table. Note that it is not limited to this, and the accuracy rate shown in Figure 10(f) can also be used as a substitute. The improved taste profile is the new taste profile obtained by the improvement process, and the improved undiscovered information is the new undiscovered information obtained by the improvement process. The difference in improvement rate from the previous time is the difference in accuracy rate. The difference in improvement content from the previous time is the difference in taste profile and undiscovered information.

[0033] The Improvement History DB300 stores improved taste profiles and unimproved information as the inference and improvement processes are repeatedly performed. In this disclosure, the improved taste profiles and unimproved information include an importance item. This importance indicates how important the taste profile and unimproved information are to the taste profile and unimproved information. A higher number indicates greater importance. This importance is referenced when calculating the prediction confidence score, which will be described later.

[0034] Here, we will explain in detail the inference process by the learning prediction ranking generation unit 103 (two-choice prediction, accuracy rate / prediction ranking calculation) and the improvement process by the improvement processing unit 105.

[0035] First, let's explain the inference process. The learning prediction ranking generation unit 103 generates a two-choice prediction prompt using a predefined phrase stored in the memory unit (not shown) for two-choice prediction, sends it to the generating AI 200, and obtains the answer result for that two-choice prediction. Figure 9 shows a specific example of a two-choice prediction prompt. As shown in the figure, this prompt includes a taste profile and target products, and is instruction information to cause the generating AI 200 to decide which of the target products to select, which one is preferred, or which one has higher sales, according to the tastes described in the taste profile. The instruction phrase here is a predefined phrase and is stored in the memory unit. The target products are the products shown in the sales ranking. In this case, they are the two selected bento boxes from the bento menu.

[0036] In Figure 9, the two-choice prediction prompt includes instructions to allow the generating AI200 to choose between A. Nori Bento and B. Chicken Nanban Bento, according to the taste profile.

[0037] The learning prediction ranking generation unit 103 then performs this operation for all combinations of products shown in the sales ranking to generate a prediction ranking.

[0038] Figure 10 is a schematic diagram illustrating a two-choice process. Figure 10(a) shows the first process (correct example), which involves selecting two products (nori bento and chicken nanban bento) from the sales ranking. The generating AI200 responds with nori bento as the bento that should be selected.

[0039] Figure 10(b) shows a summary table and counters. The summary table is a table that summarizes the number of selections for each product in the sales ranking. The counter is a table that shows the summary value indicating whether the product was selected correctly or incorrectly. This summary table and counters are repeatedly created and updated. In this case, according to the actual sales ranking, the nori bento is ranked higher than the chicken nanban bento, so it is judged to be a correct decision and 1 is added to the correct answer counter.

[0040] Figures 10(c) and 10(d) show the two-choice prediction process from the sales ranking for the second round of two-choice predictions, and the aggregation process. Figure 10(c) shows that the two choices for the two-choice prediction are nori bento and karaage bento. Here, karaage bento is selected, and as shown in Figure 10(d), 1 is added to the number of selections for karaage bento. The counter is increased by 1 for incorrect choices. This is because, in the actual sales ranking, nori bento has higher sales than karaage bento, so the generation AI200's judgment that karaage bento should be selected is incorrect.

[0041] This process is repeated. Figures 10(e) and 10(f) show the final combinations. Figure 10(g) shows the predicted ranking with the summary table sorted by the number of selections. As shown in the figure, this predicted ranking is a table that includes the correct rank (based on sales ranking), predicted rank, product name, and number of selections.

[0042] The accuracy rate calculation unit 104 is responsible for calculating an accuracy rate table (different from the accuracy rate shown in Figure 10(f)) for each product in comparison with other products, based on the predicted ranking generated by the learning prediction ranking generation unit 103. Figure 11 shows this accuracy rate table. As shown in the figure, this accuracy rate table includes an accuracy rate for each product from 1st to 10th place, indicating whether its ranking is appropriate or not when compared with each product of the other ranks.

[0043] The method for calculating this accuracy rate table will now be explained. Figure 12 shows the method for calculating the accuracy rate table. Figure 12(a) shows an overview of the process in which the generating AI 200 is made to decide which of two products selected from the prediction ranking (Figure 10(g)) is preferred. For example, as shown in the figure, the following two products are selected, and the following questions are asked to the generating AI 200. That is, the accuracy rate calculation unit 104 generates an accuracy rate calculation prompt and sends it to the generating AI 200. This accuracy rate calculation prompt includes a taste profile, as seen in the two-choice prediction prompt (including unexplored information) (see Figure 9 or Figure 15). Then, one of the following questions is included in this prompt, and the generating AI 200 is made to answer it. The order of the target products in the question affects the generating AI 200's judgment, so a different way of asking the question with a different order of choices (1 or 2 below) is randomly selected. 1. Which do you prefer: Nori Bento (1st place) or Chicken Nanban Bento (2nd place)? 2. Which do you prefer: Chicken Nanban Bento (2nd place) or Nori Bento (1st place)? In the diagram, the generating AI200 responds that it likes nori bento (seaweed bento).

[0044] This process is performed for all combinations of products shown in the prediction ranking, and this is repeated 50 times. Figure 12(b) shows the result of performing this process 50 times. Note that 50 is just one example; it could be more or fewer times.

[0045] Figure 13 shows an example of the summary of prediction results over 50 trials and a table of correct answers by pair. As shown in Figure 13(a), this example of the summary of prediction results includes the pair number, pair name, selected product for each question, frequency of correct product selection, and correct answer rate by pair. The pair name is the combination of two products selected from the products. In the figure, the product with the higher prediction ranking is listed first as the pair name. The frequency of correct product selection indicates how often the correct product was selected. The correct answer rate by pair is the value obtained by dividing that selection frequency by 50 (total number of selections). Figure 13(b) is a table of correct answers by pair that organizes the above values. For example, in Figure 13(b), comparing 1st place: Nori Bento and 2nd place: Chicken Nanban Bento, the percentage of people who selected Nori Bento (i.e., the correct answer rate) is 52%.

[0046] In this way, a two-choice selection is made and a predicted ranking is generated, which differs slightly from the actual sales ranking. Figure 14(a) shows the predicted sales ranking, and Figure 14(b) shows the actual sales ranking. As shown in the figures, the predicted sales ranking has a similar ordering to the actual sales ranking, but it is slightly different. The learning prediction ranking generation unit 103 analyzes why there is a difference between the two rankings, acquires undiscovered information, and improves the taste profile.

[0047] In the improvement process performed by the improvement processing unit 105, described later, improvement analysis is conducted, undiscovered information is added and updated, and the taste profile is improved and updated.

[0048] After the first processing S104 through S108 is completed, unexplored information is generated, so in subsequent binary predictions, the binary prediction is performed using a binary prediction prompt that includes the unexplored information. Figure 15 shows the improved taste profile and the binary prediction prompt that includes the unexplored information, in addition to the taste profile. As shown in the figure, unexplored information is added below the taste profile, and binary prediction is performed based on this. This unexplored information includes the user's preferences or behavioral tendencies other than the taste profile, as well as their importance.

[0049] Next, the improvement process will be described. The learning prediction ranking generation unit 103 generates improvement analysis prompts for analyzing improvement content based on the prediction ranking and accuracy rate. Figure 16 shows a specific example of an improvement analysis prompt. As shown in the figure, the improvement analysis prompt includes prerequisite information regarding the prediction ranking, generation instructions that instruct improvements to the taste profile and unexplored information, output instructions for intermediate information for that purpose, and output instructions for final information.

[0050] This improvement analysis prompt includes, as background information, the prompt used for the two-choice ranking prediction (i.e., the two-choice prediction prompt mentioned above), the accuracy rate, the predicted ranking, and the correct answer ranking (corresponding to the sales ranking). This information indicates the target for improvement. Among this background information, phrases such as "You performed..." and "As a result, you obtained..." are standard phrases and are stored in memory beforehand. The background information is generated by combining these standard phrases with information such as the two-choice prediction prompt.

[0051] Furthermore, the improvement analysis prompt includes generation instructions for performing the following processes. These are also standard phrases and are information stored in advance in a memory unit (not shown). Specifically, the improvement analysis is performed in the following four steps. 1. Gain a detailed understanding of the taste profiles and unexplored information that contribute to the improvement / decrease in accuracy of binary predictions. =Taste profiles that contributed to past accuracy improvements / decreases 2. Compare the prediction results with the correct answers and conduct a detailed analysis of prediction errors. 3. Identifying trends and patterns revealed by the analysis 4. Based on the content of section 3, acquire new taste profiles and undiscovered information. First, in process 1, the contents of the improvement history DB300 are referenced. Therefore, the generation instruction includes the contents of the improvement history DB. Then, by executing this process step, the following intermediate information is obtained. The [format] specification here indicates the format in which it is written. It is a pre-stored standard phrase.

[0052] <Intermediate Information> Understanding the taste profiles and unexplored information that contributed to past improvements / decreases in accuracy: <Describe the information obtained> - Improved accuracy - Taste profile - Unexplored information - Reduced accuracy - Taste profile - Unexplored information Prediction Error Analysis: <Analyzes predictions with errors to identify trends necessary to achieve a 100% accuracy rate in two-choice predictions.> - Error: - Product name: <Product name with error> - Correct Answer Ranking: <Ranking based on correct answers> - Predicted Rank: <Predicted Ranking> - Rank difference: <Difference between predicted ranking and correct ranking> - Details of the error: - Analysis of the cause of the error: - Conclusion Based on the above intermediate information, the following final information is obtained. This information is also obtained by combining it with pre-stored standard phrases.

[0053] <Final Information> Novel Taste Profile: <New taste information based on prediction error analysis> New and Untapped Information: <New and untapped information (elements other than taste) based on prediction error analysis> Improved taste profile: <Taste profiles that contributed to past accuracy improvements and new taste profiles> Improved unexplored information: <Unexplored information that contributed to past accuracy improvements and new unexplored information (elements other than taste)> In this way, the improved taste profile and improved unexplored information are generated by the generating AI 200, and these taste profiles and unexplored information are stored and updated in the prompt generation information storage unit 101. This update process is repeated until the accuracy rate in the prediction ranking remains high. Specifically, it is performed by Early Stopping until there is no longer any prospect of the average accuracy rate in the accuracy rate table improving. In this disclosure, if the accuracy rate does not improve for three consecutive times, the inference process and improvement process are terminated there, and the unexplored information and taste process with the highest accuracy rate are stored in the prompt generation information storage unit 101 as optimal information. The improved unexplored information and improved taste process are stored in the improvement history DB 300, and the unexplored information and taste profile in the column with the highest accuracy rate are retrieved and stored in the prompt generation information storage unit 101.

[0054] In the operational phase processing, product ranking predictions are made using unexplored information and taste processes that have been stored as optimal information.

[0055] Next, the test phase processing in this disclosure will be described. This test phase processing is a process to verify whether the optimal unexplored information and optimal taste profile generated in the learning phase processing are valid. In this test phase processing, sales rankings of products in different fields, products in different stores, etc., are used. For example, while information on bento shops was used in the learning phase, information on other stores may be used in the test phase.

[0056] Figure 17 shows the functional configuration of a prediction system 100 for performing the test phase processing of the present disclosure. As shown in the figure, the prediction system 100 includes a prompt generation information storage unit 101, a sales ranking acquisition unit 102, a test prediction ranking generation unit 103a, an accuracy rate calculation unit 104, and a prediction confidence level calculation unit 106.

[0057] The prompt generation information storage unit 101 and the sales ranking acquisition unit 102 are the same as described above. The test prediction ranking generation unit 103a has the same function as the training prediction ranking generation unit 103. That is, the test prediction ranking generation unit 103a performs a two-choice prediction and generates a prediction ranking (see Figure 12).

[0058] The accuracy rate calculation unit 104 has the same function as the accuracy rate calculation unit in the learning phase. That is, the accuracy rate calculation unit 104 calculates the accuracy rate for each product in comparison with other products using two-choice prediction and generates an accuracy rate table.

[0059] The prediction confidence calculation unit 106 is responsible for calculating the validity, or confidence level, of the prediction ranking generated by the test prediction ranking generation unit 103a. This prediction confidence calculation unit 106 generates a prompt for generating a prediction confidence level and causes the generating AI 200 to calculate that confidence level. The prompt generated here consists of a fixed part (stored in the memory unit in advance) and information obtained during the learning phase.

[0060] Figure 18 shows the prompts (including outlier detection) for generating prediction confidence scores. These prompts include instructions for answering the prediction confidence score for the prediction ranking, the reason for the prediction, the numerical range of the prediction confidence score, excluded products, and the reason for the exclusion. The target portion of these prompts includes the learning-optimized prompts (corresponding to undiscovered information and taste profiles), training data, and test data. The training data corresponds to the product names and product information used during training. The test data corresponds to the product names and product information of the prediction targets in the test phase (product rankings from stores and fields different from the training phase). The output format shows the prediction confidence score, reason for the exclusion, and outlier products and their reasons for exclusion for each product. The output format shows the response format from the generating AI200. This information consists of information obtained in the training phase and standardized information stored in the memory unit.

[0061] Figure 19 shows the output results of the prediction confidence score generated by the generating AI200 in accordance with the prompt for generating prediction confidence score. As shown in the figure, these output results include the product name, prediction confidence score, and the reason, in order of prediction ranking.

[0062] Figure 20 shows the output results of the outlier product generated by the AI200 in accordance with the prompt for generating prediction confidence. As shown in the figure, the outlier product and the reason for it are specifically described.

[0063] Figure 21 is a flowchart showing the operation of the prediction system 100 that performs the test phase processing.

[0064] The test prediction ranking generation unit 103a obtains prompt generation information from the prompt generation information storage unit 101 (S201) and generates a prediction ranking through a two-choice prediction process. The accuracy rate calculation unit 104 calculates the accuracy rate for each product based on the prediction ranking and generates an accuracy rate table (S203).

[0065] The prediction confidence calculation unit 106 calculates the prediction confidence level based on the sales ranking and the prediction ranking (S204) and extracts outliers (S205). The accuracy rate calculation unit 104 removes the outliers from the prediction ranking and calculates the accuracy rate for them (S206).

[0066] In this way, the test phase processing allows us to determine the accuracy rate without outlier removal (S203) and the accuracy rate with outlier removal (S206). The accuracy rate without outlier removal is calculated using the same method as the accuracy rate in the accuracy rate table (Figure 13) above. The accuracy rate with outlier removal is calculated by excluding products that have been identified as outliers and using the same calculation method as the accuracy rate without outlier removal.

[0067] Figure 22 shows the detailed accuracy tables for both with and without outlier removal. Figure 23 shows the accuracy tables for both with and without outlier removal. These accuracy tables are obtained by the following calculation.

[0068] Without outlier removal: Calculate the average accuracy rate for each pair of predicted products. When outlier removal is performed: Exclude outlier items and similarly calculate the average correct answer rate for each pair. Figure 22 is a table showing the accuracy rates for two-choice predictions. As shown in the figure, we can obtain two accuracy rates: one with outlier removal and one with outlier removal.

[0069] Operators can use this accuracy table to understand how accurately they were able to predict the target product in a two-choice selection between the target product and another product. The testing phase measures performance on unknown data. For example, if an accuracy of 70% or higher is achieved in both the learning and testing phases, it can be determined that the unexplored information and taste profiles obtained in the learning phase are highly versatile and can be used in the operational phase as well.

[0070] In other words, the validity of the prediction ranking can be determined based on these accuracy rates. The judgment may be made subjectively by the operator who requested the generation of this prediction ranking, or a threshold for the accuracy rate (for example, 70%) may be set, and if even one accuracy rate falls below the threshold, the prediction ranking, unexplored information, and taste profile may be deemed invalid. Alternatively, the judgment may be made automatically by a machine, rather than by an operator. For example, this prediction system 100 may include a judgment unit, and if the respective accuracy rates are above a threshold, the judgment unit may determine that the unexplored information and taste profile are highly versatile and optimal information, and notify or present this to the operator.

[0071] The purpose of identifying outliers is to see the results when products that differ from the learned trends are excluded during the learning phase. For example, suppose that during the learning phase, undiscovered information and taste profiles were obtained using the ranking of "sweets." Then, in the test and operation phases, if "noodles" become the target of prediction, the learned taste profiles and undiscovered information do not include information on "noodles," making prediction difficult. Therefore, different trends and categories from the learning phase may influence the results as noise, and it is possible to determine the appropriate accuracy rate after removing them.

[0072] Next, we will describe the processing involved in actual operation using the unexplored information and taste profiles obtained in the learning and testing phases. Figure 24 is a block diagram showing the functional configuration of the prediction system 100 in the operation phase. As shown in the figure, in the operation phase, the prediction system 100 consists of a prompt generation information storage unit 101, a sales ranking acquisition unit 102, an operation-use prediction ranking generation unit 103b, and a prediction confidence calculation unit 105a. The products targeted here are not limited to those targeted in the learning and testing phases.

[0073] The prompt generation information storage unit 101 stores prompt generation information (product name, product information, unopened information, and taste profile) that was deemed optimal during the learning phase. In addition, it may also store prompt generation information for which the predicted ranking was deemed appropriate during the testing phase.

[0074] The operational prediction ranking generation unit 103b is the part that generates a predicted ranking of product sales according to the same processing as the training prediction ranking generation unit 103, etc. In addition, the operational prediction ranking generation unit 103b also generates predicted rankings with and without outlier processing.

[0075] The prediction confidence calculation unit 106a is the part that calculates the confidence level (validity of the ranking) for the ranking of products described in the prediction ranking based on the sales ranking and the prediction ranking. The prediction confidence calculation unit 106a performs the same processing as the prediction confidence calculation unit 106.

[0076] Figure 25 is a flowchart illustrating the operation of the prediction system 100 during the operational phase processing. As shown in the figure, the operational prediction ranking generation unit 103b acquires prompt generation information from the prompt generation information storage unit 101 (S301). The sales ranking acquisition unit 102 acquires sales rankings from the POS system 400 (S302). Based on the sales rankings and prompt generation information, the operational prediction ranking generation unit 103b performs a two-choice prediction process and generates a prediction ranking (S303). The two-choice prediction process and the prediction ranking generation process are the same as those described in the learning phase and test phase.

[0077] The prediction confidence calculation unit 106a calculates the confidence level for the prediction ranking based on the prompt generation information and sales ranking (S304) and extracts outliers (S305). Then, the operational prediction ranking generation unit 103b generates a prediction ranking excluding the outliers (S306).

[0078] In this way, we can obtain both a predicted ranking that does not consider outliers and a predicted ranking that does consider outliers.

[0079] Figure 26 shows the predicted ranking without considering outliers and the predicted ranking with outliers considered. As shown in the figure, the prediction confidence calculation unit 106a generates a list in which products corresponding to outliers are excluded when outlier processing is enabled. In this disclosure, the shaded areas (products F and J) are the excluded products.

[0080] Next, the effects of the prediction system 100 of this disclosure will be described. This prediction system 100 executes at least one of three processing phases: a learning phase, a testing phase, and an operation phase. In this disclosure, the prediction system 100 automatically generates various prompts and causes the generating AI 200 to respond. The generated prompts are generated by automatically combining a fixed portion stored in a memory unit (not shown) with information acquired or generated by the prediction system 100.

[0081] The prediction system 100 that executes the test phase includes a prompt generation information storage unit 101 that stores prompt generation information, including the product names of multiple products to be predicted, product information, and taste profiles (preference profiles) used to generate a predicted ranking of the products.

[0082] The sales ranking acquisition unit 102 (functioning as the correct answer ranking acquisition unit) acquires the sales ranking (correct answer ranking) of products. The learning prediction ranking generation unit 103 generates a learning prediction ranking based on the selection tendencies (two-choice predictions) of multiple products generated by the generating AI 200 using prompt generation information.

[0083] The accuracy rate calculation unit 104 then calculates the accuracy rate based on the results of comparing the training prediction ranking with the sales ranking. In other words, the accuracy rate calculation unit 104 causes the generating AI 200 to determine which of the two products listed in the prediction ranking has higher sales. The accuracy rate calculation unit 104 then calculates the accuracy rate for each product in the prediction ranking based on that determination and generates an accuracy rate table.

[0084] The improvement processing unit 105 functions as a prompt information acquisition unit and, based on sales rankings, learning prediction rankings, and accuracy rates, has the generation AI 200 generate and acquire previously undiscovered information other than taste (preference) used to generate learning prediction rankings for products, as well as improved taste profiles, and stores them in the prompt generation information storage unit 101.

[0085] This process allows us to obtain unexplored information to be used in the prediction ranking.

[0086] The prompt generation information storage unit 101 then stores the unexplored information generated by the generation AI 200 as prompt generation information. The improvement processing unit 105 uses the prompt generation information to repeatedly cause the generation AI 200 to generate new unexplored information and improved taste profiles, stores them in the prompt generation information storage unit 101, and updates the prompt generation information.

[0087] Here, the improvement processing unit 105 generates an improvement analysis prompt indicating a request for improved taste profiles and undiscovered information based on the correct answer ranking, the learning prediction ranking, and the accuracy rate, and sends it to the generating AI 200, from which it obtains the improved taste profiles and undiscovered information.

[0088] The improvement processing unit 105 repeats the process based on the accuracy rate and then terminates. For example, when the accuracy rate no longer surpasses the highest value, the improvement processing unit 105 stops generating unexplored information and taste profiles and stores the unexplored information and taste profiles from the time when the accuracy rate was highest in the prompt generation information storage unit 101.

[0089] This process allows for the generation of appropriate unexplored information and taste profiles.

[0090] Furthermore, in this disclosure, the learning prediction ranking generation unit 103 causes the generating AI 200 to determine the selection tendency that indicates which of two products selected from multiple products included in the sales ranking (correct answer ranking) should be selected, based on the taste profile, and obtains the result of that determination. The learning prediction ranking generation unit 103 obtains the determination result for all combinations of multiple products and generates a learning prediction ranking based on the determination results for all multiple products.

[0091] Furthermore, the learning prediction ranking generation unit 103 generates prompts that include preference profiles, product combinations, and requests for responses regarding which product to select, and obtains judgment results from the generating AI.

[0092] This allows for the generation of a training prediction ranking based on the selection trend analysis results for all combinations of two products included in the sales ranking. This analysis can be further refined by obtaining it from the generating AI200.

[0093] Using the taste profiles and undiscovered information generated by the prediction system 100 operating in the learning phase, the prediction system 100 operating in the operational phase includes an operational prediction ranking generation unit 103b that generates test prediction rankings for multiple other products, including other categories.

[0094] This configuration allows for the generation of predictive rankings using improved taste profiles obtained during the learning phase and newly acquired unexplored information. These predictive rankings, based on unexplored information, represent appropriate rankings.

[0095] Furthermore, using the taste profiles and unexplored information generated by the prediction system 100 operating in the learning phase, the test prediction ranking generation unit 103a of the prediction system 100 operating in the test phase generates test rankings and accuracy rates for multiple products. Then, the prediction confidence calculation unit 106, which operates as an incorrect product acquisition unit, uses the taste profiles, unexplored information, and correct rankings to acquire incorrect products that deviate from the prediction confidence and learning trends. Finally, the accuracy rate calculation unit 104 calculates the accuracy rate by excluding incorrect products from the products based on the test rankings.

[0096] The prediction confidence calculation unit 106 then functions as an output unit (e.g., a communication unit or display unit) and outputs the accuracy rate for testing and the accuracy rate obtained after excluding incorrect items. This output is sent to an operator, etc. By looking at these accuracy rates, the operator can judge the validity of the prediction ranking, as well as the validity of the preference profile and undiscovered information on which it is based. For example, if both of these accuracy rates are 70% or higher, the preference profile and undiscovered information may be judged to be valid.

[0097] The apparatus and method of this disclosure have the following configurations.

[0098] [1] A storage unit that stores prompt generation information including the product names of multiple products to be predicted, product information, and preference profiles used to generate a predicted ranking of the said products, The correct answer ranking acquisition unit obtains the correct answer ranking of products, A learning prediction ranking generation unit generates a learning prediction ranking based on the selection trends of the multiple products by the generating AI generated using the prompt generation information, A unit that calculates the accuracy rate of the learning prediction ranking by comparing it with the aforementioned correct answer ranking, A prompt information acquisition unit that, based on the aforementioned correct answer ranking, the learning prediction ranking, and the accuracy rate, causes the generating AI to generate, acquire, and store in the memory unit previously unknown information other than preferences used to generate the learning prediction ranking of the products, and the improved preference profile. A device equipped with the following features.

[0099] [2] The memory unit stores the unopened information as the prompt generation information. The prompt information acquisition unit updates the storage unit by repeatedly causing the generating AI to generate the unexplored information and the preference profile using the prompt generation information. The apparatus described in [1].

[0100] [3] The prompt information acquisition unit, Based on the aforementioned correct answer ranking, the learning prediction ranking, and the accuracy rate, an improvement analysis prompt indicating a request for improvement of the preference profile and the previously undiscovered information is generated and sent to the generating AI, and the improved preference profile and the previously undiscovered information are obtained from the generating AI. The apparatus described in [1] or [2].

[0101] [4] The prompt information acquisition unit, Based on the accuracy rate, the iteration is terminated. The apparatus described in [2].

[0102] [5] The prompt information acquisition unit, When the accuracy rate no longer exceeds its highest value, the generation of the unexplored information and the preference profile is stopped, and the unexplored information and the preference profile at the time when the accuracy rate was at its highest are stored in the memory unit. The apparatus described in [4].

[0103] [6] The aforementioned learning prediction ranking generation unit is: Based on the preference profile, the result of the selection tendency judgment is obtained, which of the two products selected from the multiple products included in the correct answer ranking should be selected. The aforementioned judgment results are obtained for all combinations of multiple products. Based on the judgment results for all of the aforementioned products, a learning prediction ranking is generated. The apparatus described in any one of [1] to [5].

[0104] [7] The aforementioned learning prediction ranking generation unit is: A prompt is generated and sent to the generating AI, which includes the preference profile, the combination of products, and a request for a response regarding which of the products to select, and the decision result is obtained from the generating AI. The apparatus described in [6].

[0105] [8] [1] A ranking generation unit that generates a prediction ranking of multiple other products using the preference profile and unopened information generated in the apparatus according to claim 1, A prediction device equipped with the following features.

[0106] [9] A test ranking generation unit generates a test ranking and accuracy rate for the multiple products using the preference profile and undiscovered information generated in the apparatus described in [1], An outlier acquisition unit that acquires outlier products that deviate from the prediction confidence level and learning trend using the aforementioned preference profile, the aforementioned undiscovered information, and the correct answer ranking, A calculation unit that calculates the correct answer rate by excluding incorrect items from the products based on the aforementioned test ranking, An output unit that outputs the accuracy rate of the aforementioned test ranking and the accuracy rate obtained by excluding products that are not included in the aforementioned test ranking, A test phase processing unit equipped with the following features.

[0107]

[10] A method in a device comprising a storage unit that stores prompt generation information including product names of multiple products to be predicted, product information, and preference profiles used to generate a predicted ranking of the said products, Steps to obtain the correct ranking for products, A learning prediction ranking generation step that generates a learning prediction ranking based on the selection trends of the multiple products by the generating AI generated using the prompt generation information, A step to calculate the accuracy rate by comparing the aforementioned learning prediction ranking with the aforementioned correct answer ranking, A prompt information acquisition step involves having the generating AI generate, acquire, and store in the memory unit, untapped information other than preferences used to generate the learning prediction ranking of the products, and an improved preference profile, based on the aforementioned correct answer ranking, the aforementioned learning prediction ranking, and the aforementioned accuracy rate. A method for providing this.

[0108] Incidentally, a generative AI model (generative AI 200), exemplified by LLM, is a model that, in response to a prompt containing input information, generates content according to the instructions, context, questions, and output format indicated by the prompt, and returns that content as response information. The prompt can also include input information, in which case the generative AI model generates response information targeting the input information. The generative AI model may be an interactive AI model that includes, for example, a Large Language Model (LLM) and a user interface (UI) for interaction with the user, enabling text-based or voice-based chat with the user. Examples of such generative AI models include ChatGPT, GPT(registered trademark)-3.5, GPT-4V, PaLM2, etc. In this embodiment, the prediction system 100 enables the provision of content using multiple types of interactive AI models. These interactive AI models may be stored within the prediction system 100, or they may be stored in other devices connected to the prediction system 100 via a network, and configured to allow information exchange with the user via the prediction system 100. Although only one prediction system 100 is shown in the diagram, it may include multiple prediction systems 100.

[0109] In this disclosure, a prompt refers to information indicating instructions or questions entered by a user in an interactive system such as a dialogue with a generated AI model or a command-line interface (CLI).

[0110] The block diagram used in the description of the above embodiment shows functional units. These functional blocks (components) are realized by any combination of at least one of hardware and software. Furthermore, the method of realizing each functional block is not particularly limited. That is, each functional block may be realized using one device that is physically or logically coupled, or it may be realized using two or more physically or logically separated devices that are directly or indirectly connected (for example, using wired or wireless connections). A functional block may be realized by combining the above one device or the above multiple devices with software.

[0111] Functions include, but are not limited to, judgment, decision, judgment, calculation, calculation, processing, derivation, investigation, exploration, confirmation, reception, transmission, output, access, resolution, selection, selection, establishment, comparison, assumption, expectation, assumption, broadcasting, notifying, communicating, forwarding, configuring, reconfiguring, allocating (mapping), and assigning. For example, a functional block (configuration part) that enables transmission is called a transmitting unit or transmitter. As mentioned above, the method of implementation is not particularly limited.

[0112] For example, the prediction system 100 in one embodiment of the present disclosure may function as a computer that performs processing of the information processing method of the present disclosure. Figure 27 is a diagram showing an example of the hardware configuration of the prediction system 100 according to one embodiment of the present disclosure. The prediction system 100 described above may be physically configured as a computer device including a processor 1001, memory 1002, storage 1003, communication device 1004, input device 1005, output device 1006, bus 1007, etc.

[0113] In the following explanation, the term "device" can be replaced with "circuit," "device," "unit," etc. The hardware configuration of the prediction system 100 may include one or more of the devices shown in the figure, or it may be configured to omit some of the devices.

[0114] Each function in the prediction system 100 is realized by loading predetermined software (programs) onto hardware such as the processor 1001 and memory 1002, which allows the processor 1001 to perform calculations, control communication by the communication device 1004, and control at least one of data reading and writing in the memory 1002 and storage 1003.

[0115] The processor 1001 controls the entire computer, for example, by running the operating system. The processor 1001 may be composed of a central processing unit (CPU) that includes interfaces with peripheral devices, control devices, arithmetic units, registers, etc. For example, the above-mentioned learning prediction ranking generation unit 103, operation prediction ranking generation unit 103b, test prediction ranking generation unit 103a, accuracy rate calculation unit 104, improvement processing unit 105, and prediction confidence calculation unit 106 may be implemented by the processor 1001.

[0116] Furthermore, the processor 1001 reads programs (program code), software modules, data, etc., from at least one of the storage 1003 and the communication device 1004 into the memory 1002 and executes various processes accordingly. The program used is one that causes the computer to execute at least a part of the operations described in the above embodiment. For example, the learning prediction ranking generation unit 103, the operation prediction ranking generation unit 103b, the test prediction ranking generation unit 103a, the accuracy rate calculation unit 104, the improvement processing unit 105, and the prediction confidence calculation unit 106 may be implemented by a control program stored in the memory 1002 and running on the processor 1001, and other functional blocks may be implemented similarly. The above-described processes have been explained as being executed by one processor 1001, but they may be executed simultaneously or sequentially by two or more processors 1001. The processor 1001 may be implemented by one or more chips. The program may also be transmitted from a network via a telecommunications line.

[0117] Memory 1002 is a computer-readable recording medium and may consist of at least one of the following: ROM (Read Only Memory), EPROM (Erasable Programmable ROM), EEPROM (Electrically Erasable Programmable ROM), RAM (Random Access Memory), etc. Memory 1002 may also be called a register, cache, main memory, etc. Memory 1002 can store executable programs (program code), software modules, etc., for carrying out an information processing method according to one embodiment of the present disclosure.

[0118] Storage 1003 is a computer-readable recording medium and may consist of at least one of the following: an optical disc such as a CD-ROM (Compact Disc ROM), a hard disk drive, a flexible disk, a magneto-optical disk (e.g., a compact disc, a digital multipurpose disc, a Blu-ray® disc), a smart card, flash memory (e.g., a card, a stick, a key drive), a floppy® disk, a magnetic strip, etc. Storage 1003 may also be called an auxiliary storage device. The above-mentioned storage medium may be, for example, a database, server, or other suitable medium including at least one of memory 1002 and storage 1003.

[0119] The communication device 1004 is hardware (transmitting / receiving device) for communicating between computers via at least one of a wired network and a wireless network, and is also referred to as a network device, network controller, network card, communication module, etc. The communication device 1004 may be configured to include high-frequency switches, duplexers, filters, frequency synthesizers, etc., in order to implement at least one of frequency division duplex (FDD) and time division duplex (TDD). For example, the sales ranking acquisition unit 102 described above may be implemented by the communication device 1004. The communication device 1004 may be implemented with physically or logically separated transmitting and receiving units.

[0120] The input device 1005 is an input device that accepts input from an external source (e.g., a keyboard, mouse, microphone, switch, button, sensor, etc.). The output device 1006 is an output device that outputs to an external source (e.g., a display, speaker, LED lamp, etc.). The input device 1005 and the output device 1006 may be configured as an integrated unit (e.g., a touch panel).

[0121] Furthermore, each device, such as the processor 1001 and memory 1002, is connected by a bus 1007 for communicating information. The bus 1007 may be configured using a single bus, or different buses may be configured for each device.

[0122] Furthermore, the prediction system 100 may also include hardware such as a microprocessor, a digital signal processor (DSP), an ASIC (Application Specific Integrated Circuit), a PLD (Programmable Logic Device), and an FPGA (Field Programmable Gate Array), and some or all of each functional block may be realized by such hardware. For example, the processor 1001 may be implemented using at least one of these hardware components.

[0123] Information notification is not limited to the embodiments described herein and may be carried out by other means. For example, information notification may be carried out by physical layer signaling (e.g., DCI (Downlink Control Information), UCI (Uplink Control Information)), upper layer signaling (e.g., RRC (Radio Resource Control) signaling, MAC (Medium Access Control) signaling, broadcast information (MIB (Master Information Block), SIB (System Information Block))), other signals, or combinations thereof. RRC signaling may also be called RRC messages, and may be, for example, RRC Connection Setup messages, RRC Connection Reconfiguration messages, etc.

[0124] The processing procedures, sequences, flowcharts, etc., of each aspect / embodiment described herein may be reordered, provided they are consistent with each other. For example, the methods described herein present various step elements in an exemplary order and are not limited to that specific order.

[0125] Input and output information may be stored in a specific location (e.g., memory) or managed using a management table. Input and output information may be overwritten, updated, or appended to. Output information may be deleted. Input information may be transmitted to other devices.

[0126] The determination may be made by a value represented by 1 bit (0 or 1), by a boolean value (true or false), or by a numerical comparison (for example, a comparison with a predetermined value).

[0127] Each aspect / embodiment described herein may be used individually, in combination, or switched between as needed during implementation. Furthermore, notification of specific information (e.g., notification that "X is") is not limited to explicit notification, but may also be implicit (e.g., by not providing such notification).

[0128] Although the present disclosure has been described in detail above, it will be clear to those skilled in the art that the present disclosure is not limited to the embodiments described herein. The present disclosure can be implemented in modified and altered forms without departing from the intent and scope of the present disclosure as defined by the claims. Accordingly, the descriptions in the present disclosure are illustrative and not intended to be restrictive in any way.

[0129] Software should be broadly interpreted to mean instructions, instruction sets, code, code segments, program code, programs, subprograms, software modules, applications, software applications, software packages, routines, subroutines, objects, executable files, execution threads, procedures, functions, and so on, whether they are called software, firmware, middleware, microcode, hardware description languages, or by any other name.

[0130] Furthermore, software, instructions, information, etc., may be transmitted and received via a transmission medium. For example, if software is transmitted from a website, server, or other remote source using at least one of wired technologies (such as coaxial cable, fiber optic cable, twisted pair, or digital subscriber line (DSL)) and wireless technologies (such as infrared or microwave), then at least one of these wired and wireless technologies is included in the definition of a transmission medium.

[0131] The information, signals, etc. described in this disclosure may be represented using any of the various different techniques. For example, the data, instructions, commands, information, signals, bits, symbols, chips, etc. that may be referred to throughout the above description may be represented by voltage, current, electromagnetic waves, magnetic fields or magnetic particles, optical fields or photons, or any combination thereof.

[0132] In addition, terms used in this disclosure and terms necessary for understanding this disclosure may be replaced with terms having the same or similar meanings. For example, at least one of the channel and symbol may be a signal (signaling). Also, a signal may be a message. Furthermore, a component carrier (CC) may be called a carrier frequency, cell, frequency carrier, etc.

[0133] Furthermore, the information, parameters, etc., described in this disclosure may be expressed using absolute values, relative values ​​from a given value, or other corresponding information. For example, wireless resources may be indicated by an index.

[0134] The names used for the parameters described above are not restrictive in any way. Furthermore, the formulas and other expressions using these parameters may differ from those expressly disclosed in this disclosure. Various channels (e.g., PUCCH, PDCCH, etc.) and information elements can be identified by any suitable name, and therefore, the various names assigned to these various channels and information elements are not restrictive in any way.

[0135] In this disclosure, terms such as "Mobile Station (MS)," "user terminal," "User Equipment (UE)," and "terminal" may be used interchangeably.

[0136] A mobile station may also be referred to by those skilled in the art as a subscriber station, mobile unit, subscriber unit, wireless unit, remote unit, mobile device, wireless device, wireless communication device, remote device, mobile subscriber station, access terminal, mobile terminal, wireless terminal, remote terminal, handset, user agent, mobile client, client, or some other appropriate term.

[0137] As used in this disclosure, the terms “determining” and “determining” may encompass a wide variety of actions. “Determining” may include, for example, judging, calculating, computing, processing, deriving, investigating, looking up, searching, or inquiring (e.g., searching in a table, database, or other data structure), or ascertaining. “Determining” may also include, for example, receiving (e.g., receiving information), transmitting (e.g., sending information), inputting, outputting, or accessing (e.g., accessing data in memory). Furthermore, "judgment" and "decision" can include considering something as having been "judged" or "decided" after resolving, selecting, choosing, establishing, comparing, etc. In other words, "judgment" and "decision" can include considering something as having been "judged" or "decided" after some action. Also, "judgment (decision)" can be reinterpreted as "assuming," "expecting," or "considering."

[0138] The terms “connected,” “coupled,” or any variation thereof, mean any direct or indirect connection or coupling between two or more elements, and may include the presence of one or more intermediate elements between two elements that are “connected” or “coupled” with each other. The coupling or connection between elements may be physical, logical, or a combination thereof. For example, “connection” may be reinterpreted as “access.” As used in this disclosure, two elements may be considered to be “connected” or “coupled” with each other using at least one of one or more wires, cables, and printed electrical connections, and, in some non-limiting and non-exclusive examples, electromagnetic energy having wavelengths in the radio frequency domain, microwave domain, and optical (both visible and invisible) domain.

[0139] In this disclosure, the phrase "based on" does not mean "based solely on" unless otherwise specified. In other words, the phrase "based on" means both "based solely on" and "based at least on."

[0140] Any reference to elements using designations such as “first,” “second,” etc., as used in this disclosure does not generally limit the quantity or order of those elements. These designations may be used in this disclosure as a convenient way to distinguish between two or more elements. Accordingly, references to first and second elements do not imply that only two elements may be adopted, or that the first element must precede the second element in any way.

[0141] Where the terms “include,” “including,” and their variations are used in this disclosure, these terms are intended to be inclusive, as is the term “comprising.” Furthermore, the term “or” as used in this disclosure is not intended to mean exclusive OR.

[0142] In this disclosure, if articles are added by translation, such as a, an, and the in English, this disclosure may include the fact that the noun following these articles is plural.

[0143] In this disclosure, the term "A and B are different" may mean "A and B are different from each other." The term may also mean "A and B are each different from C." Terms such as "separate" and "combine" may be interpreted similarly to "different." [Explanation of Symbols]

[0144] 100...Prediction system, 100a...Processing unit, 101...Prompt generation information storage unit, 200...Generating AI, 300...Improvement history DB, 400...POS system, 500...Sales ranking data, 600...Questionnaire, 102...Sales ranking acquisition unit, 103...Training prediction ranking generation unit, 103a...Test prediction ranking generation unit, 103b...Operation prediction ranking generation unit, 104...Accuracy rate calculation unit, 105...Improvement processing unit, 105a...Prediction confidence level calculation unit, 106...Prediction confidence level calculation unit.

Claims

1. A storage unit that stores prompt generation information including the product names of multiple products to be predicted, product information, and preference profiles used to generate a sales prediction ranking of the said products, The correct ranking acquisition unit obtains the sales ranking of products, A learning prediction ranking generation unit generates a learning prediction ranking based on the selection trends of the multiple products generated by the generating AI using the prompt generation information, A unit that calculates the accuracy rate of the learning prediction ranking by comparing it with the aforementioned correct answer ranking, A prompt information acquisition unit that, based on the aforementioned correct answer ranking, the aforementioned learning prediction ranking, and the aforementioned accuracy rate, causes the generating AI to generate, acquire, and store in the memory unit previously unknown information other than preferences used to generate the learning prediction ranking of the products, and the improved preference profile. A device equipped with the following features.

2. The memory unit stores the unopened information as the prompt generation information. The prompt information acquisition unit uses the prompt generation information to cause the generating AI to repeatedly generate the unexplored information and the preference profile, thereby updating the storage unit. The apparatus according to claim 1.

3. The prompt information acquisition unit, Based on the correct answer ranking, the learning prediction ranking, and the accuracy rate, an improvement analysis prompt indicating a request for improvement of the preference profile and the undiscovered information is generated and sent to the generating AI, and the improved preference profile and the undiscovered information are obtained from the generating AI. The apparatus according to claim 1.

4. The prompt information acquisition unit, Based on the accuracy rate, the iteration is terminated. The apparatus according to claim 2.

5. The prompt information acquisition unit, When the accuracy rate no longer exceeds its highest value, the generation of the unexplored information and the preference profile is stopped, and the unexplored information and the preference profile at the time when the accuracy rate was at its highest are stored in the memory unit. The apparatus according to claim 4.

6. The aforementioned learning prediction ranking generation unit is: Based on the aforementioned preference profile, the result of the selection tendency judgment is obtained, which of the two products selected from the multiple products included in the correct answer ranking should be selected. The aforementioned judgment results are obtained for all combinations of multiple products. Based on the judgment results for all of the aforementioned products, a learning prediction ranking is generated. The apparatus according to claim 1.

7. The aforementioned learning prediction ranking generation unit is: A prompt including the preference profile, the combination of products, and a request for a response regarding which of the products to select is generated and sent to the generating AI, and the decision result is obtained from the generating AI. The apparatus according to claim 6.

8. A ranking generation unit that generates a predicted ranking of multiple other products using the preference profile and undiscovered information generated in the apparatus according to claim 1, A prediction device equipped with the following features.

9. A test ranking generation unit that generates a test ranking and accuracy rate for the plurality of products using the preference profile and undiscovered information generated in the apparatus according to claim 1, An outlier acquisition unit that acquires outlier products that deviate from the prediction confidence level and learning trend using the aforementioned preference profile, the aforementioned undiscovered information, and the correct answer ranking, A calculation unit that calculates the correct answer rate by excluding incorrect items from the products based on the aforementioned test ranking, An output unit that outputs the accuracy rate of the aforementioned test ranking and the accuracy rate obtained by excluding products that are not included in the aforementioned test ranking, A test phase processing unit equipped with the following features.

10. A method in a device comprising a storage unit that stores prompt generation information including product names of multiple products to be predicted, product information, and preference profiles used to generate a predicted ranking of the said products, Steps to obtain the correct ranking for products, A learning prediction ranking generation step that generates a learning prediction ranking based on the selection trends of the multiple products by the generating AI generated using the prompt generation information, A step to calculate the accuracy rate by comparing the aforementioned learning prediction ranking with the aforementioned correct answer ranking, A prompt information acquisition step involves having the generating AI generate, acquire, and store in the memory unit, untapped information other than preferences used to generate the learning prediction ranking of the product, and an improved preference profile, based on the aforementioned correct answer ranking, the aforementioned learning prediction ranking, and the aforementioned accuracy rate. A method for providing this.

Images