Information processing device, information processing method, and program
The system improves name matching accuracy by using a pre-trained model to perform multiple tasks concurrently, addressing the high learning costs and reduced accuracy issues of previous methods.
Patent Information
- Application Number
- JP2024028010
- Authority / Receiving Office
- JP · JP
- Patent Type
- Applications
- Current Assignee / Owner
- Filing Date
- 2024-02-28
- Publication Date
- 2025-07-08
AI Technical Summary
Existing methods require retraining a machine learning model for each name matching task, leading to high learning costs and reduced accuracy.
A system that uses a pre-trained model to perform name matching and title inference tasks simultaneously, leveraging two separate learning datasets to improve accuracy while maintaining versatility.
Enhances determination accuracy in name matching tasks without the need for task-specific retraining, maintaining model generality.
Smart Images
Figure 2025102599000001_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to an information processing apparatus, an information processing method, and a program.
Background Art
[0002] There is known a technique for determining name matching using a machine learning model obtained by fine-tuning a pre-trained language model (for example, Non-Patent Document 1).
Prior Art Documents
Non-Patent Documents
[0003]
Non-Patent Document 1
Summary of the Invention
Problems to be Solved by the Invention
[0004] Depending on existing methods, it is necessary to train a machine learning model for each task or name matching target in order to improve accuracy, and retraining is required when performing name matching for other tasks or name matching targets, resulting in a large learning cost.
[0005] Therefore, the present invention has been made in view of these points, and an object thereof is to improve the determination accuracy of the grouping task while maintaining versatility.
Means for Solving the Problems
[0006] In the information processing apparatus according to the first aspect of the present invention, when inference data in which the name of the first grouping target, the name of the second grouping target, the description of the first grouping target, and the description of the second grouping target are associated is input, a learned model that outputs information indicating whether the first grouping target in the input inference data matches the second grouping target in the inference data, (1) Based on the first learning dataset in which the name of the first grouping target, the name of the second grouping target, the description of the first grouping target, the description of the second grouping target, and a label indicating whether the first grouping target and the second grouping target match are associated, when the name of the first grouping target, the name of the second grouping target, the description of the first grouping target, and the description of the second grouping target are input, a grouping task that outputs information indicating whether the first grouping target and the second grouping target match, and (2) a title inference task that outputs the name of the estimation target using the description of the estimation target as an input based on the second learning dataset in which the name of the estimation target and the description of the estimation target are associated, a storage unit that stores the learned model that has learned the above, an acquisition unit that acquires the inference data, an input unit that inputs the acquired inference data into the learned model, and determines whether the first grouping target in the inference data matches the second grouping target in the inference data. And an output unit that outputs the result determined by the determination unit.
[0007] In the description of the first naming target associated in the inference data and the first learning dataset, it includes information indicating details of the first naming target which is a product, and the amount of the first naming target. In the description of the second naming target associated in the inference data and the first learning dataset, it includes information indicating details of the second naming target which is a product, and the amount of the second naming target. In the description of the estimation target associated in the second learning dataset, it may include information indicating details of the estimation target.
[0008] The learned model may learn the naming task based on the first learning dataset for a general language model, and the title inference task based on the second learning dataset.
[0009] The acquisition unit acquires the first learning dataset and the second learning dataset. The information processing apparatus further has a learning unit that generates the learned model that has learned (1) the naming task based on the first learning dataset acquired by the acquisition unit, and (2) the title inference task based on the second learning dataset, and stores the learned learned model in the storage unit.
[0010] The learning unit may learn the naming task based on the first learning dataset and the title inference task based on the second learning dataset in parallel in a single learning process.
[0011] In the information processing method according to the second aspect of the present invention, an acquisition unit that acquires inference data to be executed by a computer, the inference data being associated with the name of a first naming target, the name of a second naming target, the description of the first naming target, and the description of the second naming target; and a learned model stored in a storage unit that outputs information indicating whether or not the first naming target in the inference data matches the second naming target in the inference data. The learned model is learned based on: (1) a first learning dataset in which the name of the first naming target, the name of the second naming target, the description of the first naming target, the description of the second naming target, and a label indicating whether or not the first naming target matches the second naming target are associated, and which outputs information indicating whether or not the first naming target matches the second naming target when the name of the first naming target, the name of the second naming target, the description of the first naming target, and the description of the second naming target are input (a naming task); and (2) a second learning dataset in which the name of an estimation target and the description of the estimation target are associated, and which outputs the name of the estimation target with the description of the estimation target as input (a title inference task). The method includes a step of inputting the inference data into the learned model and determining whether or not the first naming target in the inference data matches the second naming target in the inference data, and a step of outputting the result determined in the determining step.
[0012] In the program according to the third aspect of the present invention, the computer is caused to perform: an acquisition unit that acquires inference data in which the name of a first naming target, the name of a second naming target, the description of the first naming target, and the description of the second naming target are associated; and a learned model stored in a storage unit, which outputs information indicating whether or not the first naming target in the inference data matches the second naming target in the inference data when the inference data is input. The learned model is learned based on: (1) a first learning dataset in which the name of a first naming target, the name of a second naming target, the description of the first naming target, the description of the second naming target, and a label indicating whether or not the first naming target matches the second naming target are associated, and which outputs information indicating whether or not the first naming target matches the second naming target when the name of the first naming target, the name of the second naming target, the description of the first naming target, and the description of the second naming target are input; and (2) a second learning dataset in which the name of an estimation target and the description of the estimation target are associated, and which outputs the name of the estimation target when the description of the estimation target is input. The learned model is caused to input the inference data and determine whether or not the first naming target in the inference data matches the second naming target in the inference data, and output the result determined in the determining step.
Advantages of the Invention
[0013] According to the present invention, it is possible to improve the determination accuracy of the naming task while maintaining versatility.
Brief Description of the Drawings
[0014]
Figure 1
Figure 2
Figure 3
Figure 4
Figure 5
Figure 6
Embodiments for Carrying Out the Invention
[0015] [Overview of the Information Processing System S] FIG. 1 is a diagram for explaining the overview of the information processing system S. FIG. 1(a) shows the configuration of the information processing system S. The information processing system S is a system for performing name matching. Name matching is a task executed by a machine learning model and is a task for determining whether a plurality of given objects match.
[0016] The objects to be name-matched by the information processing apparatus system S are, as an example, names of products or services, but are not limited thereto. The information processing system S may perform name matching on corporate names, personal names, or other names. The information processing system S includes an information processing apparatus 1 and an information terminal 2. The information processing apparatus 1 and the information terminal 2 are communicably connected via a network.
[0017] The information processing apparatus 1 is an apparatus for performing name matching. The information processing apparatus 1 is, as an example, a server. The information processing apparatus 1 trains a machine learning model and, when data of name-matching targets is given, determines whether the targets in the given data match using the machine learning model.
[0018] The information terminal 2 is a terminal used by a user of the information processing system S. The information terminal 2, as an example, transmits a dataset used for learning or inference to the information processing apparatus 1, instructs the information processing apparatus 1 to execute learning or inference, receives an inference result from the information processing apparatus 1, and causes it to be displayed on a display unit. Note that the information processing apparatus 1 and the information terminal 2 may be integrally configured. That is, the information processing apparatus 1 is provided with an input / output interface, accepts operations from a user, and displays inference results.
[0019] Referring to FIG. 1(b), the processing in the information processing system S will be described. The information processing apparatus 1 stores a pre-trained model M1. The pre-trained model M1 is a general-purpose language model, which is a trained model that has been trained to be able to execute natural language processing tasks based on a large amount of datasets. The information processing apparatus 1 trains the pre-trained model M1 with a name matching task and a title inference task to generate a trained model M2.
[0020] The name matching task is a task in which a plurality of names of name matching targets to be name-matched and texts indicating explanations for each name matching target are given, and it is determined whether the given plurality of name matching targets match. The name of the name matching target indicates the name of the product, natural person, corporation, etc. to be name-matched. The explanation of the name matching target indicates the nature of the name matching target. For example, when the name matching target is a product, the explanation of the name matching target includes the size, color, function, manufacturing location, manufacturer, seller, model number, operating environment, raw materials, selling points, price, etc. of the product.
[0021] In addition, when the name matching target is a natural person, the explanation of the name matching target includes information such as date of birth, place of birth, alma mater, occupation, achievements, etc. When the name matching target is a corporation, the explanation of the name matching target includes information such as the address of the corporation, history such as the number of employees and establishment year, composition of officers, sales amount, etc.
[0022] Specifically, the information processing apparatus 1 trains the pre-trained model M1 with a name matching task based on the first training dataset. An example of the first training dataset is shown in FIG. 2(a). In the first training dataset, the name of the first name matching target, the name of the second name matching target, the explanation of the first name matching target, the explanation of the second name matching target, and a label indicating whether the first name matching target and the second name matching target match are associated.
[0023] The title inference task is a task of generating the title of the object to be estimated from the text showing the explanation about the object to be estimated. Specifically, the information processing apparatus 1 causes the pre-trained model M1 to learn the title inference task based on the second learning dataset. An example of the second learning dataset is shown in Fig. 2(b). In the second learning dataset, the name of the object to be estimated and the explanation of the object to be estimated are associated with each other.
[0024] In the second learning dataset, it is particularly preferable to perform learning based on a dataset including various product names, model numbers, brand names, place names, or corporate names, etc. as specific expressions as explanations or titles. By learning the title inference task, the pre-trained model M2 can learn specific expressions that can affect the title and are used in the explanation. As a result, the model can recognize important expressions in the text that affect the result of name alignment. As a result, it is possible to expect the effect that the accuracy of the name alignment task is improved without impairing the generality of the model.
[0025] The pre-trained model M2 is learned to output a determination result D2 corresponding to the input inference data D1 when the inference data D1 is input. The inference data D1 is associated with the name of the first name alignment target, the name of the second name alignment target, the explanation of the first name alignment target, and the explanation of the second name alignment target. The determination result D2 indicates whether or not the first name alignment target in the inference data matches the second name alignment target in the inference data.
[0026] The information processing apparatus 1 inputs the inference data D1 to the pre-trained model M2 and causes it to output the determination result D2.
[0027] By configuring the information processing system S in this way, there is an effect that the determination accuracy of the name alignment task can be improved while maintaining generality.
[0028] [Configuration of Information Processing Apparatus 1] FIG. 3 is a block diagram showing the configuration of the information processing apparatus 1. The information processing apparatus 1 includes a communication unit 11, a storage unit 12, and a control unit 13. The control unit 13 includes an acquisition unit 131, a determination unit 132, an output unit 133, and a learning unit 134.
[0029] The communication unit 11 is a communication interface for transmitting and receiving data to and from other devices via a network. The storage unit 12 is a storage medium including a ROM (Read Only Memory), a RAM (Random Access Memory), an SSD (Solid State Drive), a hard disk drive, and the like. The storage unit 12 stores in advance a program executed by the control unit 13. The storage unit 12 stores a pre-trained model M1 and a trained model M2.
[0030] The control unit 13 is a processor such as a CPU (Central Processing Unit). The control unit 13 functions as the acquisition unit 131, the determination unit 132, the output unit 133, and the learning unit 134 by executing a program stored in the storage unit 12.
[0031] The acquisition unit 131 acquires inference data D1. As an example, the acquisition unit 131 acquires the inference data D1 from the information terminal 2. The acquisition unit 131 may acquire the inference data D1 from the storage unit 12 or may acquire it from an external device (not shown). The acquisition unit 131 may acquire a first learning dataset and a second learning dataset and output them to the learning unit 134.
[0032] The determination unit 132 inputs the acquired inference data into the trained model M2 and determines whether or not a first clustering target in the inference data matches a second clustering target in the inference data. The output unit 133 outputs a determination result D2 determined by the determination unit 132. As an example, the output unit 133 causes the display unit of the information terminal 2 to display the determination result D2.
[0033] When the target to be clustered by the information processing apparatus 1 is a product, the clustering may be performed based on a dataset including the price of the product.
[0034] In the description of the first naming target associated in the inference data and the first learning dataset, it includes information indicating the details of the first naming target that is a product and the amount of the first naming target. In the description of the second naming target associated in the inference data and the first learning dataset, it includes information indicating the details of the second naming target that is a product and the amount of the second naming target.
[0035] Note that even when using a dataset including the amount of a product in the naming task, it is not necessary to use the amount of the product for learning in the title inference task. That is, in the description of the estimation target associated in the second learning dataset, it includes information indicating the details of the estimation target. This is because while the amount of the product has a great influence on the result of the naming task, the influence of the amount of the product on the result of the title inference task is relatively small.
[0036] By performing naming based on the information including the amount of the product to be named in this way, the accuracy of naming can be improved.
[0037] The learning unit 134 learns the pre-trained model M1 based on the first learning dataset and the second learning dataset acquired by the acquisition unit 131, generates a learned model M2 by updating the parameters of the pre-trained model M1, and stores the generated learned model M2 in the storage unit 12. Note that the learning unit 134 may additionally train the learned model M2 with a naming task or a title inference task.
[0038] Specifically, the learning unit 134 trains a pre-trained model M1 based on the first learning dataset. Specifically, the learning unit 134 inputs the name of the first clustering target, the name of the second clustering target, the description of the first clustering target, and the description of the second clustering target into the pre-trained model M1, and outputs a determination result indicating whether the first clustering target and the second clustering target match. The learning unit 134 calculates a loss based on the determination result output by the pre-trained model M1 and the label included in the first learning dataset, and updates the parameters of the pre-trained model M1 based on the calculated loss to train the pre-trained model M1.
[0039] The learning unit 134 inputs the description of the estimation target included in the second learning dataset into the pre-trained model M1, and outputs the name of the estimation target corresponding to the input description. The learning unit 134 calculates a loss based on the name of the estimation target output by the pre-trained model M1 and the name of the estimation target as the teacher data included in the second learning dataset, and updates the parameters of the pre-trained model M1 based on the calculated loss to train the pre-trained model M1.
[0040] The learning of the clustering task and the learning of the title inference task may be executed simultaneously. The learning unit 134 may concurrently learn the clustering task based on the first learning dataset and the title inference task based on the second learning dataset in a single learning process. FIG. 4 is a flowchart for explaining the learning flow in this case. The flowchart shown in FIG. 4 starts from the point when the information processing apparatus 1 acquires an instruction to start learning from the information terminal 2.
[0041] The acquisition unit 131 acquires the first learning dataset (S01). The acquisition unit 131 acquires the second learning dataset (S02). The learning unit 134 determines an end condition (S03). The end condition is, for example, that a predetermined number of learning sessions have been performed.
[0042] When the end condition is not satisfied (NO in S03), the learning unit 134 causes the pre-trained model M1 to perform the clustering task based on the first learning dataset and outputs the result (S04). The learning unit 134 causes the pre-trained model M1 to perform the title inference task based on the second learning dataset and outputs the result (S05).
[0043] The learning unit 134 calculates the loss based on the result output by the pre-trained model M1 in the clustering task and the label associated in the first learning dataset (S06). Also, the learning unit 134 calculates the loss based on the result output by the pre-trained model M1 in the title inference task and the name to be estimated as the teacher data associated in the second learning dataset (S06).
[0044] The learning unit 134 updates the parameters of the pre-trained model M1 based on the calculated loss (S07). As an example, the learning unit 134 calculates the gradient in the title inference task and the gradient in the clustering task based on the calculated loss, and updates the parameters based on the average value of the gradient in the title inference task and the gradient in the clustering task.
[0045] Note that how much the parameters are updated per step may be different between the title inference task and the clustering task. That is, the update amount of the parameters may be calculated by multiplying each gradient by a different predetermined coefficient, or different learning rates may be set for the title inference task and the clustering task. The information processing apparatus 1 proceeds with the process to S03.
[0046] When the end condition is satisfied (YES in S03), the learned model M2, which is the pre-trained model M1 with the parameter update completed, is stored in the storage unit 12 (S08). Then, the information processing apparatus 1 ends the process.
[0047] The acquisition unit 131 may be configured to acquire inference data included in a prompt (instruction) described in a natural language. As an example, the acquisition unit 131 displays a screen for receiving a prompt on the information terminal 2, and acquires a prompt described in a natural language from the information terminal 2. FIG. 5 shows an example of a prompt acquired by the acquisition unit 131. As shown in FIG. 5, the prompt includes an instruction (P1) for the content of the task to be executed, a first naming target (P2), and a second naming target (P3). The first naming target (P2) and the second naming target each include a name (P21, P31) and an explanation (P22 and P32).
[0048] In this case, the learned model M2 is trained to take the prompt as an input, identify the content of the task to be executed included in the prompt, and execute the naming task based on the inference data included in the prompt when the content of the identified task is a naming task.
[0049] [Flow of processing in the information processing apparatus 1] FIG. 6 is a flowchart showing the flow of processing in the information processing apparatus 1. The flowchart shown in FIG. 6 starts when an instruction to perform inference is received from the information terminal 2.
[0050] The acquisition unit 131 acquires inference data (S11). The acquisition unit 131 inputs the inference data into the learned model M2, and determines whether the first naming target in the inference data matches the second naming target in the inference data (S12). The output unit 133 outputs the determination result output by the learned model M2 (S13). As an example, the output unit 133 causes the information terminal 2 to display the determination result output by the learned model M2. Then, the information processing apparatus 1 ends the processing.
[0051] [Effects in this embodiment] As described above, by training the information processing apparatus 1 in the title inference task and the naming task, it is possible to improve the determination accuracy of the naming task while maintaining versatility without specializing in a specific task.
[0052] The present invention has been described using the embodiments. However, the technical scope of the present invention is not limited to the scope described in the above embodiments, and various modifications and changes are possible within the scope of the gist. For example, all or part of the device can be configured by functionally or physically dispersing and integrating it in any unit. Also, new embodiments resulting from any combination of a plurality of embodiments are included in the embodiments of the present invention. The effects of the new embodiments resulting from the combination have the effects of the original embodiments combined.
Explanation of Reference Numerals
[0053] 1 Information processing apparatus 2 Information terminal 11 Communication unit 12 Storage unit 13 Control unit 131 Acquisition unit 132 Determination unit 133 Output unit 134 Learning unit
Claims
1. A data set for inference, which, when input with inference data in which the name of the first naming target, the name of the second naming target, the description of the first naming target, and the description of the second naming target are associated, outputs information indicating whether the first naming target in the input inference data matches the second naming target in the inference data. The learning model is based on: (1) Based on a first training data set in which the name of the first naming target, the name of the second naming target, the description of the first naming target, the description of the second naming target, and a label indicating whether the first naming target and the second naming target match are associated, when the name of the first naming target, the name of the second naming target, the description of the first naming target, and the description of the second naming target are input, it is a naming task that outputs information indicating whether the first naming target and the second naming target match; (2) Based on a second training data set in which the name of the estimation target and the description of the estimation target are associated, a title inference task that outputs the name of the estimation target using the description of the estimation target as input; A storage unit that stores the learned model that has learned the above; An acquisition unit that acquires the inference data; An input unit that inputs the acquired inference data into the learned model and determines whether the first naming target in the inference data matches the second naming target in the inference data; An output unit that outputs the result determined by the determination unit; An information processing apparatus having the above components.
2. In the description of the first naming target associated in the inference data and the first training data set, it includes information indicating the details of the first naming target that is a product and the amount of the first naming target. In the description of the second naming target associated in the inference data and the first training data set, it includes information indicating the details of the second naming target that is a product and the amount of the second naming target. In the description of the estimation target associated in the second training data set, it includes information indicating the details of the estimation target. The information processing apparatus according to Claim 1.
3. The learned model is a learned model that has learned the naming task based on the first training data set and the title inference task based on the second training data set for a general-purpose language model. The information processing apparatus according to claim 1.
4. The acquisition unit acquires the first learning dataset and the second learning dataset, The information processing apparatus, (1) the clustering task based on the first learning dataset acquired by the acquisition unit, and (2) the title inference task based on the second learning dataset, further includes a learning unit that generates the learned model learned from the above and stores the learned model in the storage unit. The information processing apparatus according to claim 1.
5. The learning unit learns the clustering task based on the first learning dataset and the title inference task based on the second learning dataset in parallel in a single learning process. The information processing apparatus according to claim 4.
6. An acquisition unit that acquires inference data in which the name of the first clustering target, the name of the second clustering target, the description of the first clustering target, and the description of the second clustering target are associated, A learned model stored in the storage unit, which outputs information indicating whether the first clustering target in the inference data and the second clustering target in the inference data match when the inference data is input. (1) Based on the first learning dataset in which the name of the first clustering target, the name of the second clustering target, the description of the first clustering target, the description of the second clustering target, and a label indicating whether the first clustering target and the second clustering target match are associated, when the name of the first clustering target, the name of the second clustering target, the description of the first clustering target, and the description of the second clustering target are input, the clustering task that outputs information indicating whether the first clustering target and the second clustering target match, (2) Based on the second learning dataset in which the name of the estimation target and the description of the estimation target are associated, the title inference task that outputs the name of the estimation target with the description of the estimation target as the input, inputting the inference data into the learned model that has learned the above, and determining whether the first clustering target in the inference data and the second clustering target in the inference data match; outputting the result determined in the determining step; An information processing method having
7. To the computer, An acquisition unit that acquires inference data in which the name of a first naming target, the name of a second naming target, the description of the first naming target, and the description of the second naming target are associated, A learned model stored in a storage unit, which outputs information indicating whether or not a first naming target in the inference data matches a second naming target in the inference data when the inference data is input. The learned model is, (1) Based on a first learning dataset in which the name of a first naming target, the name of a second naming target, the description of the first naming target, the description of the second naming target, and a label indicating whether or not the first naming target and the second naming target match are associated, a naming task that outputs information indicating whether or not the first naming target and the second naming target match when the name of the first naming target, the name of the second naming target, the description of the first naming target, and the description of the second naming target are input, (2) A title inference task that outputs the name of the estimation target with the description of the estimation target as input based on a second learning dataset in which the name of the estimation target and the description of the estimation target are associated, Inputting the inference data into the learned model that has learned the above, and determining whether or not a first naming target in the inference data matches a second naming target in the inference data, Outputting the result determined in the determining step, A program for causing the above to be executed.