Training device, analyzing device, training method, and program

Abstract

This training device includes: an acquisition unit that acquires training data including an action sequence that is a time series of action element labels representing details of past actions for each customer; a conversion unit that converts, for each customer on the basis of the training data, the action sequence into a set sequence representing a set of a specific number of action element labels, in which null elements are used, for each time; and a training unit that uses the converted set sequence as training data to train an action model that receives input of the set sequence to output a vector representing details of an action for each time such that a score for each action element label obtained from output of the action model corresponds to the set of a specific number of action element labels included in the training data.

Description

Learning device, analysis device, learning method, and program

[0001] The technology of the present disclosure relates to a learning device, an analysis device, a learning method, and a program.

[0002] Companies are accumulating customer behavioral history at various touchpoints and analyzing the customer journey, the process of customer behavior from becoming aware of a service, signing a contract, and continuing to use the service. They are then using the results of this analysis to improve their marketing and customer support measures. For example, they are analyzing the timing at which customers are likely to open an advertisement for a new service based on their behavioral history at touchpoints from app delivery to web access, or analyzing the procedural steps that make customers more likely to need support at a call center or in-store based on their behavioral history at touchpoints from web access to a call center.

[0003] Transformer is known as a technology for analyzing customer behavior processes. When modeling with Transformer, customer behavior history is first represented as a table consisting of records containing at least three items: customer, time, and behavior label representing the behavior content. Then, a series of behavior labels arranged in chronological order for each customer is prepared as training data. This training data is loaded into Transformer, and model parameters are estimated using self-supervised learning. In self-supervised learning, the behavior label for the next time or a series in which randomly selected behavior labels are masked is input into Transformer, and the task of assigning the masked behavior label from the output using multi-class classification is repeatedly performed. BERT4Rec (Non-Patent Document 1) is an example of analyzing customer behavior history with Transformer in this way and using it for recommendations.

[0004] Fei Sun, et al. BERT4Rec: Sequential Recommendation with Bidirectional Encoder Representations from Transformer. CIKM, 2019. https: / / arxiv.org / abs / 1904.06690

[0005] Conventional technology assigned one behavior label to one behavior, making it difficult to model and analyze cases in which multiple behaviors occurred simultaneously.

[0006] In the prior art, there are two possible means for representing multiple actions that occur simultaneously.

[0007] One is a method for representing multiple simultaneous actions as a single action label. For example, if there are three action elements—"changing rate plans," "purchasing a device," and "subscribing for options"—and a combination of these action elements is included in a single action, up to eight different action labels can be assigned: no action, only changing rate plans, only purchasing a device, only subscribing for options, changing rate plans and purchasing a device, changing rate plans and subscribing for options, purchasing a device and subscribing for options, and changing rate plans, purchasing a device, and subscribing for options. This method has the problem that the number of labels increases as the number of combinations increases. As the number of embedding vectors increases, the computational load for learning and inference increases.

[0008] The other method is to arrange multiple simultaneous behaviors as if they occurred in a sequence. For example, when behaviors A and B occur simultaneously, they are treated as behavior labels arranged in the order of behavior A → behavior B. This method increases the length of the sequence and the amount of calculation. Also, even if the behaviors occurred simultaneously, the model will overfit even if they are arranged in a hypothetical order.

[0009] The disclosed technology has been made in consideration of the above points, and aims to provide a learning device, an analysis device, a learning method, and a program that can analyze behavior while reducing computational costs, even when multiple behavioral contents occur simultaneously.

[0010] A first aspect of the present disclosure is a learning device that includes: an acquisition unit that acquires learning data including an action series that is a time series of action element labels that represent past action details for each customer; a conversion unit that converts, based on the learning data, for each customer, into a set series that represents a set consisting of a predetermined number of action element labels using null elements for each time point; and a learning unit that uses the converted set series as training data to learn a behavior model that takes the set series as input and outputs a vector that represents action details for each time point, such that a score for each action element label obtained from the output of the behavior model corresponds to a set consisting of the predetermined number of action element labels included in the training data.

[0011] A second aspect of the present disclosure is an analysis device including: an acquisition unit that acquires a behavior series, which is a time series of behavior element labels that represent past behavioral content of a customer; a conversion unit that converts, based on the behavior series, into a set series that represents a set consisting of a predetermined number of behavior element labels using null elements for each time; and an analysis unit that uses a trained behavior model that receives the set series as input and outputs a vector that represents the behavior content for each time when the converted set series is input, and that uses the trained behavior model to output a vector that represents the behavior content for each time, wherein the behavior model is trained in advance using the converted set series as training data so that a score for each behavior element label obtained from the output of the behavior model corresponds to a set consisting of the predetermined number of behavior element labels included in the training data.

[0012] A third aspect of the present disclosure is a learning method, in which a computer acquires learning data including a behavioral sequence, which is a time series of behavioral element labels representing past behavioral details for each customer, converts the learning data into a set sequence representing a set consisting of a predetermined number of behavioral element labels using null elements for each time point for each customer, based on the learning data, and uses the converted set sequence as training data to learn a behavioral model that takes the set sequence as input and outputs a vector representing behavioral details for each time point, such that a score for each behavioral element label obtained from the output of the behavioral model corresponds to the set consisting of the predetermined number of behavioral element labels included in the training data.

[0013] A fourth aspect of the present disclosure is a program for causing a computer to function as the learning device of the first aspect or the analysis device of the second aspect.

[0014] According to the disclosed technology, even when multiple behavioral details occur simultaneously, the behavior can be analyzed while suppressing calculation costs.

[0015] 1 is a schematic block diagram of an example of a computer that functions as a learning device and an analysis device of this embodiment. FIG. 2 is a diagram showing an example of a past behavior sequence. FIG. 3 is a block diagram showing the functional configuration of the learning device of this embodiment. FIG. 4 is a diagram showing an example of a set sequence. FIG. 5 is a diagram showing an example of a dictionary of behavior element labels. FIG. 6 is a diagram showing an example of a set sequence converted by a dictionary. FIG. 7 is a diagram showing an example of a design of a layer of a behavior model. FIG. 8 is a diagram showing an example of an embedding layer of a behavior model. FIG. 9 is a diagram for explaining a learning method for a behavior model. FIG. 10 is a block diagram showing the functional configuration of the analysis device of this embodiment. FIG. 11 is a flowchart showing the flow of a learning process of this embodiment. FIG. 12 is a flowchart showing the flow of an analysis process of this embodiment.

[0016] An example of an embodiment of the disclosed technology will be described below with reference to the drawings. Note that the same or equivalent components and parts in each drawing are given the same reference numerals. Also, the dimensional proportions in the drawings are exaggerated for the convenience of explanation and may differ from the actual proportions.

[0017] <Configuration of Learning Device According to This Embodiment> FIG. 1 is a block diagram showing the hardware configuration of a learning device 10 according to this embodiment.

[0018] 1, the learning device 10 includes a CPU (Central Processing Unit) 11, a ROM (Read Only Memory) 12, a RAM 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other via a bus 19 so as to be able to communicate with each other.

[0019] The CPU 11 is a central processing unit that executes various programs and controls each component. That is, the CPU 11 reads programs from the ROM 12 or the storage 14 and executes the programs using the RAM 13 as a work area. The CPU 11 controls the above components and performs various arithmetic processing in accordance with the programs stored in the ROM 12 or the storage 14. In this embodiment, a learning program is stored in the ROM 12 or the storage 14. The learning program may be a single program, or a group of programs consisting of multiple programs or modules.

[0020] The ROM 12 stores various programs and various data. The RAM 13 temporarily stores programs or data as a working area. The storage 14 is configured with an HDD (Hard Disk Drive) or an SSD (Solid State Drive) and stores various programs including the operating system and various data.

[0021] The input unit 15 includes a pointing device such as a mouse and a keyboard, and is used to input various data including learning data including behavioral sequences, which are time series of behavioral element labels that represent the past behavioral details of each customer.

[0022] For example, as shown in FIG. 2, a past behavior sequence prepared for each customer is input to the input unit 15. FIG. 2 shows an example in which a past behavior sequence is composed of records including at least three items: customer, time, and a set of behavior element labels. In this example, the first record indicates that customer A's behavior of "browsing the web - pricing plan explanation site" was observed at 11:32 on 2023 / 01 / 24. Among these, the second and third records are the only records in which two or more behavior element labels were observed. The second record indicates that customer A's two behaviors of "call center inquiry - check current pricing plan" and "call center inquiry - make a store visit appointment" were simultaneously observed at 11:40 on 2023 / 01 / 24. The third record indicates that three actions were observed simultaneously for customer A on 2023 / 01 / 24 15:03: "Store_Procedure_Change_Price_Purchase_Device", and "Store_Procedure_Change_Payment_Credit_Payment".

[0023] The display unit 16 is, for example, a liquid crystal display, and displays various information including the processing results. The display unit 16 may be a touch panel type and function as the input unit 15.

[0024] The communication interface 17 is an interface for communicating with other devices, and uses standards such as Ethernet (registered trademark), FDDI, and Wi-Fi (registered trademark).

[0025] Next, a description will be given of the functional configuration of the learning device 10. Fig. 3 is a block diagram showing an example of the functional configuration of the learning device 10.

[0026] As shown in FIG. 3, the learning device 10 functionally includes an acquisition unit 20, a conversion unit 21, a learning unit 22, and a model storage unit 24.

[0027] The acquisition unit 20 acquires the input learning data.

[0028] The conversion unit 21 converts the learning data into a set sequence representing a set consisting of a predetermined number of behavior element labels using null elements for each customer at each time point, based on the learning data.

[0029] For example, the behavioral sequence of customer A shown in FIG. 2 is converted into a set sequence representing a set of a predetermined number of behavioral element labels using empty elements for each time, as shown in FIG.

[0030] In the example of Fig. 4, the predetermined number, which is the number of elements included in the set of behavior element labels, is set to the number that maximizes the number of behavior element labels (3 in the example of Fig. 2). Empty elements that do not have a behavior element label are distinguished from others by filling them with the special behavior element "[PAD]".

[0031] The learning unit 22 uses the converted set sequence as training data to learn a behavioral model that takes the set sequence as input and outputs a vector representing the behavior content for each time point so that the score for each behavioral element label obtained from the output of the behavioral model corresponds to a set consisting of a predetermined number of behavioral element labels included in the training data.

[0032] Specifically, the learning unit 22 uses the converted set sequence as training data to learn the parameters of the behavior model 100 through self-supervised learning using binary classification for each behavior element label.

[0033] More specifically, first, a dictionary of behavior element labels is created using the set sequence converted by the conversion unit 21 as training data. Figure 5 shows an example of a behavior element label dictionary. The dictionary consists of two columns: behavior element ID and behavior element label. The behavior element IDs "0", "1", and "2" are set to the special behavior elements "[PAD]", ""[MASK]", and ""[UNK]", respectively, representing empty elements, masked elements, and behavior elements not registered in the dictionary. From behavior element ID "3" onwards, behavior elements observed in the training data are registered in order. Note that the registration order of these behavior elements is in descending order of frequency, and less frequent behavior elements may be omitted and treated as "[UNK]".

[0034] Next, the behavior element labels included in the set sequence of the training data are converted into behavior element IDs according to the dictionary. Figure 6 shows an example of a set sequence converted using the dictionary. By converting the behavior element labels into behavior element IDs, the format becomes one that can be converted using the Embedding function.

[0035] The set sequence converted using the dictionary is then input to the behavior model.

[0036] As shown in FIG. 7, the behavioral model 100 has a Transformer encoder, which receives as input a set of behavioral element IDs x1, x2, ... at each time included in a set sequence, and outputs vectors y1, y2, ... representing the behavior content at each time.

[0037] The behavioral model 100 also includes an Embedding layer, a Self-Attension layer, a Layer Norm layer, a Feed Forward layer, a Layer Norm layer, and a Linear layer.

[0038] FIG. 8 shows the details of the Embedding layer extracted from the behavioral model 100.

[0039] This example shows a state in which three behavior element IDs at a certain time in the set sequence are input to the Embedding layer. In the Time Embedding layer, a time embedding function is used to convert the time into an n-dimensional vector. n can be 2 or greater, and in this embodiment, n = 512. In the Token Embedding layer, an embedding function assigned to each behavior element ID is used to convert the behavior element ID into an n-dimensional vector. The sum of the converted n-dimensional vectors is the output of the Embedding layer. The Embedding function holds a matrix W composed of n-dimensional vectors equal to the number of types of behavior element IDs, and when a behavior element ID is input, it outputs the corresponding n-dimensional vector.

[0040] The output y1, y2, ... yn of the behavior model 100 is an n-dimensional vector. The dot product of this n-dimensional vector and matrix W is calculated to obtain a scalar value equal to the number of behavior element IDs. This scalar value is regarded as a binary classification score for each behavior element ID and compared with the correct label to obtain a cross-entropy error. Model parameters that minimize this error are estimated. Note that this is not limited to cross-entropy error, and weighted cross-entropy error may also be used as long as it is possible to calculate the error in binary classification.

[0041] FIG. 9 is an illustration of how the correct label and error are calculated for y3, which is output when a set of behavioral element IDs at a certain time in the set sequence is input to the behavioral model 100 and the inner product with the matrix W is converted to a value between 0 and 1 using the sigmoid function.

[0042] In the prior art, the softmax function is used instead of the sigmoid function, and the correct label is set so that only one behavioral element ID is 1 and the others are 0, and the error is calculated using multi-class classification that compares all behavioral element IDs at once.

[0043] Furthermore, the behavior model 100 is not limited to a Transformer encoder, but may be anything that can accept a past behavior sequence and output a vector representing the behavior content, such as a multi-layer neural network or gradient boosting.

[0044] Furthermore, in the self-supervised learning of the behavioral model 100, the set sequence input to the behavioral model 100 may be masked, and the parameters may be repeatedly updated so that the masked portions can be output from the behavioral model 100.

[0045] Furthermore, the model for analyzing behavior, which includes the behavioral model 100, which is the encoder of the Transformer, and a decoder, is trained in two stages: in the first stage, self-supervised training of the above-mentioned behavioral model 100 is performed, and in the second stage, the entire model for analyzing behavior is trained.

[0046] The learning unit 22 stores the learned behavioral model 100 in the model storage unit 24 .

[0047] <Configuration of Analysis Apparatus According to This Embodiment> FIG. 1 is a block diagram showing the hardware configuration of an analysis apparatus 50 according to this embodiment.

[0048] 1, like the learning device 10, the analysis device 50 has a CPU 11, a ROM 12, a RAM 13, a storage 14, an input unit 15, a display unit 16, and a communication interface (I / F) 17. Each component is connected to each other so as to be able to communicate with each other via a bus 19. An analysis program is stored in the ROM 12 or the storage 14.

[0049] The input unit 15 is used to input various information including a behavior sequence, which is a time series of behavior element labels that represent the past behavior of the customer to be analyzed.

[0050] Specifically, a behavior sequence is composed of a record that includes at least three items: a customer, a time, and a set of behavior element labels.

[0051] Next, a description will be given of the functional configuration of the analysis device 50. Fig. 10 is a block diagram showing an example of the functional configuration of the analysis device 50.

[0052] As shown in FIG. 10, the analysis device 50 functionally includes a model storage unit 52, an acquisition unit 54, a conversion unit 55, and an analysis unit 56.

[0053] The model storage unit 52 stores a trained behavior model 100 similar to that of the model storage unit 24 of the learning device 10 .

[0054] The acquisition unit 54 acquires the input behavioral sequence of the customer to be analyzed.

[0055] Similar to the converter 21 of the learning device 10, the converter 55 converts the action sequence into a set sequence representing a set made up of a predetermined number of action element labels using empty elements for each time, based on the action sequence.

[0056] The analysis unit 56 obtains a vector representing the behavioral content for each time when the converted set sequence is input using the trained behavior model 100. The analysis unit 56 further analyzes the behavior of the target customer by using the output of the behavior model 100 as input to a decoder included in the model for analyzing behavior.

[0057] <Operation of the Learning Device According to the Present Embodiment> Next, the operation of the learning device 10 according to the present embodiment will be described.

[0058] 11 is a flowchart showing the flow of the learning process by the learning device 10. The learning process is performed by the CPU 11 reading a learning program from the ROM 12 or storage 14, expanding it into the RAM 13, and executing it. Also, it is assumed that learning data including behavior sequences, which are time series of behavior element labels representing the past behavior of each customer, have been input to the learning device 10. The learning process is an example of a learning method.

[0059] In step S100 , the CPU 11 functions as the acquisition unit 20 to acquire a plurality of pieces of learning data received by the input unit 15 .

[0060] In step S102, the CPU 11 functions as the conversion unit 21 to convert each behavioral sequence of the plurality of learning data into a set sequence representing a set consisting of a predetermined number of behavioral element labels using empty elements for each customer and each time point.

[0061] In step S104, the CPU 11, as the learning unit 22, uses the converted set sequence as training data to learn the parameters of the behavioral model 100 through self-supervised learning using binary classification for each behavioral element label, stores the learned behavioral model 100 in the model storage unit 24, and terminates the learning process.

[0062] <Operation of the Analysis Device According to the Present Embodiment> Next, the operation of the analysis device 50 according to the present embodiment will be described.

[0063] 12 is a flowchart showing the flow of analysis processing by the analysis device 50. The analysis processing is performed by the CPU 11 reading out an analysis program from the ROM 12 or storage 14, expanding it into the RAM 13, and executing it. The model storage unit 52 of the analysis device 50 stores the learned behavior model 100 learned by the learning device 10. The analysis device 50 is also assumed to have input thereto the past behavioral patterns of the customer to be analyzed.

[0064] In step S110, the CPU 11 functions as the acquisition unit 54 to acquire the past behavioral sequence of the customer to be analyzed.

[0065] In step S112, the CPU 11 functions as the conversion unit 55 to convert the acquired behavior sequence into a set sequence representing a set made up of a predetermined number of behavior element labels using empty elements for each time.

[0066] In step S114, the CPU 11, functioning as the analysis unit 56, obtains a vector representing the behavioral content at each time when the converted set sequence is input, using the trained behavior model 100. Furthermore, the CPU 11 uses the output of the behavior model 100 as input to a decoder included in the model for analyzing behavior, analyzes the behavior of the customer being analyzed, and ends the analysis process.

[0067] As described above, the learning device according to this embodiment acquires learning data including a behavioral sequence for each customer, converts the data into a set sequence representing a set of a predetermined number of behavioral element labels using null elements for each time point for each customer, and uses the converted set sequence as training data to train a behavioral model that takes the set sequence as input and outputs a vector representing the behavioral content for each time point so that the score for each behavioral element label obtained from the output of the behavioral model corresponds to the set of the predetermined number of behavioral element labels included in the training data. This makes it possible to train a behavioral model that can analyze behavior while reducing computational costs, even when multiple behavioral content occurs simultaneously.

[0068] The analysis device according to this embodiment converts a customer's behavioral sequence into a set sequence representing a set of a predetermined number of behavioral element labels using null elements for each time, and uses a trained behavior model that takes the set sequence as input and outputs a vector representing the behavioral content for each time to calculate the vector representing the behavioral content for each time when the converted set sequence is taken as input. This makes it possible to analyze behavior while reducing calculation costs, even when multiple behavioral content occurs simultaneously.

[0069] Furthermore, the accuracy of the process analysis of customer behavior can be improved even when multiple behavioral contents occur simultaneously.

[0070] Furthermore, instead of assigning one behavior label to each behavior, we treat a set of behavior element labels that make up the behavior as a sequence. This reduces the number of label types and enables us to estimate model parameters without overfitting.

[0071] Furthermore, compared to expressing multiple simultaneous actions as a single action label, the number of label types can be reduced. This not only avoids the increase in computational complexity that occurs when the number of labels increases, but also the problem of overfitting due to frequency variations in action labels.

[0072] Furthermore, compared to arranging multiple simultaneous behaviors as if they occurred in a sequence, this method avoids the problem of overlearning, which occurs when meaningful behavior sequences are mixed with meaningless behavior sequences that are arranged provisionally.

[0073] Furthermore, for the self-supervised learning of the behavioral model, a task of assigning masked behavioral element labels using binary classification for each behavioral element label is used, rather than a task of assigning masked behavioral element labels using multi-class classification. This makes it possible to avoid overfitting to some behavioral element labels and improve the accuracy of customer behavior process analysis.

[0074] Furthermore, the present invention is not limited to the device configuration and operation of the above-described embodiment, and various modifications and applications are possible within the scope of the gist of the present invention.

[0075] For example, although the above description has been given taking the case where the analysis device and the learning device are configured separately, the present invention is not limited to this and the analysis device and the learning device may be configured as a single device.

[0076] In addition, the various processes executed by the CPU after reading the software (program) in the above embodiments may be executed by various processors other than the CPU. Examples of processors in this case include PLDs (Programmable Logic Devices) whose circuit configuration can be changed after manufacture, such as FPGAs (Field-Programmable Gate Arrays), and dedicated electrical circuits, such as ASICs (Application Specific Integrated Circuits), which are processors having a circuit configuration designed specifically to execute specific processes. Furthermore, the learning process and analysis process may be executed by one of these various processors, or by a combination of two or more processors of the same or different types (e.g., multiple FPGAs, a combination of a CPU and an FPGA, etc.). Furthermore, the hardware structure of these various processors is, more specifically, an electrical circuit that combines circuit elements such as semiconductor elements.

[0077] In addition, in each of the above embodiments, the learning program and the analysis program are described as being pre-stored (installed) in the storage 14, but this is not limiting. The programs may be provided in a form stored on a non-transitory storage medium such as a CD-ROM (Compact Disk Read Only Memory), a DVD-ROM (Digital Versatile Disk Read Only Memory), or a USB (Universal Serial Bus) memory. The programs may also be downloaded from an external device via a network.

[0078] The following additional notes are provided regarding the above-described embodiments.

[0079] (Supplementary Item 1) A learning device comprising: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire learning data including an action series, which is a time series of action element labels representing past action details for each customer; convert, based on the learning data, for each customer, into a set series representing a set consisting of a predetermined number of action element labels using null elements for each time point; and use the converted set series as training data to learn a behavior model that takes the set series as input and outputs a vector representing action details for each time point, such that a score for each action element label obtained from the output of the behavior model corresponds to the set consisting of the predetermined number of action element labels included in the training data.

[0080] (Supplementary Item 2) A non-transitory storage medium storing a program executable by a computer to execute a learning process, wherein the learning process acquires learning data including a behavior series that is a time series of behavior element labels that represent past behavioral details for each customer, converts the learning data into a set series that represents a set consisting of a predetermined number of behavior element labels using empty elements for each time point for each customer, and uses the converted set series as training data to learn a behavior model that takes the set series as input and outputs a vector that represents the behavioral details for each time point, so that a score for each behavior element label obtained from the output of the behavior model corresponds to a set consisting of the predetermined number of behavior element labels included in the training data.

[0081] (Supplementary Item 3) An analysis device including: a memory; and at least one processor connected to the memory, wherein the processor is configured to: acquire a behavior series, which is a time series of behavior element labels representing past behavioral details of a customer; convert, based on the behavior series, into a set series representing a set consisting of a predetermined number of behavior element labels using null elements for each time; and, using a trained behavior model that receives the set series as an input and outputs a vector representing the behavioral details for each time, determine a vector representing the behavioral details for each time when the converted set series is input; and the behavior model is trained in advance using the converted set series as training data so that a score for each behavior element label obtained from the output of the behavior model corresponds to a set consisting of the predetermined number of behavior element labels included in the training data.

[0082] (Supplementary Item 4) A non-transitory storage medium storing a program executable by a computer to perform an analysis process, wherein the analysis process: acquires an action series, which is a time series of action element labels representing past action details of a customer; converts the action series into a set series representing a set consisting of a predetermined number of action element labels using null elements for each time point based on the action series; uses the set series as an input and a trained action model that outputs a vector representing the action details for each time point to determine a vector representing the action details for each time point when the converted set series is input; and the action model is trained in advance using the converted set series as training data so that a score for each action element label obtained from the output of the action model corresponds to a set consisting of the predetermined number of action element labels included in the training data.

[0083] REFERENCE SIGNS LIST 10 Learning device 11 CPU 13 RAM 14 Storage 15 Input unit 16 Display unit 20, 54 Acquisition unit 21, 55 Conversion unit 22 Learning unit 24, 52 Model storage unit 50 Analysis device 56 Estimation unit 100 Behavioral model

Claims

1. An acquisition unit that acquires learning data including an action sequence that is a time series of action element labels representing past action contents for each customer; a conversion unit that, based on the learning data, converts, for each customer, into a set sequence representing a set of a predetermined number of action element labels using empty elements for each time; and a learning unit that uses the converted set sequence as training data, inputs the set sequence, and learns an action model that outputs a vector representing the action content for each time so that the score for each action element label obtained from the output of the action model corresponds to the set of the predetermined number of action element labels included in the training data. A learning device comprising:

2. The learning device according to claim 1, wherein the action model has an Embedding function using a matrix W for assigning an Embedding vector to each action element label, and the learning unit obtains the score for each action element label by taking the inner product of the output of the action model and the matrix W.

3. The learning device according to claim 1, wherein the learning unit learns the parameters of the action model by self-supervised learning using the converted set sequence as training data.

4. The learning device according to claim 1, wherein the action model has an encoder of Transformer, and the encoder inputs the set sequence.

5. An acquisition unit that acquires an action sequence that is a time series of action element labels representing a customer's past action content; a conversion unit that, based on the action sequence, converts into a set sequence representing a set of a predetermined number of action element labels using empty elements for each time; and an analysis unit that uses a learned action model that inputs the set sequence and outputs a vector representing the action content for each time to obtain a vector representing the action content for each time when the converted set sequence is input. An analysis device comprising: The action model is pre-learned so that the score for each action element label obtained from the output of the action model corresponds to the set of the predetermined number of action element labels included in the training data using the converted set sequence as training data.

6. Obtain learning data including an action sequence that is a time series of action element labels representing the past action content for each customer, convert, for each customer based on the learning data, into a set sequence representing a set of a predetermined number of action element labels using empty elements for each time point, use the converted set sequence as training data, input the set sequence, and train an action model that outputs a vector representing the action content for each time point so that the score for each action element label obtained from the output of the action model corresponds to the set consisting of the predetermined number of action element labels included in the training data. A learning method executed by a computer.

7. A program for causing a computer to function as the learning device according to any one of claims 1 to 4 or the analysis device according to claim 5.

Images