Method and system for obtaining trademark filtering and classification results through product titles
By conducting multiple iterative training and data enhancement processing on massive product title data, the problems of low efficiency and insufficient accuracy of trademark processing in the existing technology are solved, and more efficient trademark filtering and classification results are achieved, and identification accuracy is improved.
Patent Information
- Application Number
- CN202410810803.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-06-21
- Publication Date
- 2024-11-22
- Estimated Expiration
- 2044-06-21
AI Technical Summary
The prior art is inefficient and limited in trademark processing and analysis, and cannot meet current needs.
By collecting and marking massive product title data, multiple iterative training models are carried out, combining keyword extraction, key sentence extraction and data enhancement processing, the classification database of trademarks for trademarks, product titles for products and trademarks and products is trained in steps to improve training Effects and recognition accuracy.
It improves the accuracy and efficiency of trademark filter classification results, improves the accuracy of identification, and can better distinguish trademarks and product classifications.
Smart Images

Figure CN118503432B_ABST
Abstract
Description
Technical Field
[0001] The present application belongs to the field of Internet and e-commerce technology, and specifically relates to a method and system for obtaining trademark filtering and classification results through product titles. Background Art
[0002] With the development of science and technology, the application of information processing technology, trademark analysis technology and computer science is becoming more and more extensive. Among them, the processing and analysis of trademarks are important parts of these fields. The processing and analysis of these data can help us better understand and utilize these data, thereby improving our work efficiency and quality.
[0003] Existing technologies mainly process and analyze trademarks manually. Although this method can obtain certain results, it is inefficient and prone to errors. In addition, there are some technologies that process and analyze these data through traditional machine learning methods, such as support vector machines (SVMs) and decision trees. However, these methods have limited processing capabilities and accuracy and cannot meet current needs. Summary of the invention
[0004] The present application provides a method and system for obtaining trademark filtering and classification results through product titles, which can obtain trademark filtering results through product titles in e-commerce, thereby improving accuracy and efficiency.
[0005] In a first aspect, the present application provides a method for obtaining trademark filtering classification results through product titles, the method comprising the following steps:
[0006] Collect and label massive product title data, and classify the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database;
[0007] Inputting a large amount of title data as training data into the model to be trained and performing multiple rounds of iterative training to obtain a trained model. One round of iterative training specifically includes: inputting the marked data in the trademark classification database of trademark pairs into the model to be trained as the first training data group and performing the first round of training to obtain a first training model; inputting the marked data in the product classification database of product titles into the first training model as the second training data group and performing the second round of training to obtain a second training model; inputting the product classification database as the third training data group into the second training model and obtaining the third training model to complete one round of iterative training;
[0008] Obtain product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain a trademark filtering and classification result of the input data.
[0009] In a second aspect, a system for obtaining trademark filtering and classification results through product titles is provided, the system comprising:
[0010] A collection and classification unit, used to collect and mark massive product title data, and classify the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database;
[0011] A training unit is used to input a large amount of title data as training data into a model to be trained to perform multiple rounds of iterative training to obtain a trained model, wherein one round of iterative training specifically includes: inputting the marked data in the trademark classification database of trademark pairs into the model to be trained as a first training data group to perform a first round of training to obtain a first training model, inputting the marked data in the product classification database of product titles into the first training model as a second training data group to perform a second round of training to obtain a second training model, and inputting the product classification database as a third training data group into the second training model to obtain a third training model to complete one round of iterative training;
[0012] The execution unit is used to obtain product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain the trademark filtering and classification results of the input data.
[0013] In a third aspect, the present application provides a computer storage medium storing a computer program for electronic data exchange, wherein the computer program enables a computer to execute some or all of the steps described in the first aspect of the present application.
[0014] The embodiments of the present application have the following beneficial effects:
[0015] The technical solution provided by the present application collects and marks massive product title data, classifies the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database; inputs the massive title data as training data into the model to be trained to perform multiple rounds of iterative training to obtain a trained model, and a round of iterative training specifically includes: inputting the marked data in the trademark-to-trademark classification database as the first training data group into the model to be trained to perform the first round of training to obtain the first training model, inputting the marked data in the product title-to-product classification database as the second training data group into the first training model to perform the second round of training to obtain the second training model, and inputting the product classification database as the third training data group into the second training model to obtain the third training model to complete a round of iterative training. Obtain the product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain the trademark filtering classification result of the input data. The above scheme improves the training effect by performing step-by-step training on three types of databases, thereby improving the accuracy of recognition. BRIEF DESCRIPTION OF THE DRAWINGS
[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.
[0017] Figure 1 It is a structural schematic diagram of a classification and filtering system provided in an embodiment of the present application;
[0018] Figure 2 It is a flowchart of a method for obtaining trademark filtering and classification results through product titles provided in an embodiment of the present application;
[0019] Figure 3 It is a structural block diagram of a system for obtaining trademark filtering and classification results through product titles provided in an embodiment of the present application. DETAILED DESCRIPTION
[0020] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.
[0021] The terms "first", "second", etc. in the specification and claims of this application and the above-mentioned drawings are used to distinguish different objects, rather than to describe a specific order. In addition, the terms "including" and "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not limited to the listed steps or units, but optionally includes steps or units that are not listed, or optionally includes other steps or units inherent to these processes, methods, systems, products or devices.
[0022] Reference to "embodiments" herein means that a particular feature, structure, or characteristic described in conjunction with the embodiments may be included in at least one embodiment of the present application. The appearance of the phrase in various locations in the specification does not necessarily refer to the same embodiment, nor is it an independent or alternative embodiment that is mutually exclusive with other embodiments. It is explicitly and implicitly understood by those skilled in the art that the embodiments described herein may be combined with other embodiments.
[0023] The International Trademark Classification (Nice Classification) is a system designed for trademark application and management. It consists of 30 major categories, each of which has multiple subcategories, totaling more than 100 subcategories. Each subcategory has multiple subcategories, totaling nearly 5,000 goods or service items.
[0024] Locarno Chinese is the Locarno Classification (LOC) is an international classification for the registration of industrial designs, established by the Locarno Agreement (1968). The tenth edition of the Locarno Classification came into force on January 1, 2014.
[0025] The following is a schematic diagram of the architecture of a storage system involved in an embodiment of the present application.
[0026] See also Figure 1 , Figure 1 A structural diagram of a classification and filtering system provided in this application, such as Figure 1 As shown, the above classification system includes multiple terminals, each terminal includes: at least one processor and a random access memory RAM, and may also include a communication interface (Communications Interface) and a memory (such as a hard disk). Among them, the processor, RAM, communication interface and memory can all communicate with each other through a bus. The communication interface can transmit information, and the above communication interface can have a wireless communication function, and the above wireless communication function can be a short-range wireless communication or a long-range communication function (such as an NR method). The processor can call the logic instructions in the memory to execute or support the method in the embodiment of the present application.
[0027] See also Figure 2 , Figure 2A flowchart of a method for obtaining trademark filtering and classification results through product titles is provided, such as Figure 2 As shown, the above method is Figure 1 The method is performed in the classification and filtering system shown in the figure, and comprises the following steps:
[0028] Step S201, collecting and marking massive product title data, classifying the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database;
[0029] For example, the above trademark-to-trademark classification database may specifically be: a trademark-to-Nice classification database, and of course, a trademark-to-LOC classification database, and so on.
[0030] The above-mentioned product title to product classification database may specifically be: a database of product titles and e-commerce categories. Taking mobile phones as an example, such as a certain mobile phone, its corresponding product category is mobile phones, and the e-commerce category may be a category defined by the e-commerce company itself.
[0031] The above-mentioned trademark and product classification database may specifically be: a database of trademark classification and product e-commerce classification.
[0032] Step S202: input the massive title data as training data into the model to be trained to perform multiple rounds of iterative training to obtain a trained model. One round of iterative training specifically includes: inputting the marked data in the trademark-to-trademark classification database as the first training data group into the model to be trained to perform a first round of training to obtain a first training model; inputting the marked data in the product title-to-product classification database as the second training data group into the first training model to perform a second round of training to obtain a second training model; inputting the product classification database as the third training data group into the second training model to obtain a third training model to complete one round of iterative training.
[0033] The training data is divided into three databases, and each round of iterative training is trained sequentially in order to reduce the overfitting of the training model. If all the training data are mixed together for training, the data sets are random. For massive data, this training may adjust some weight values corresponding to the trademark classification data of the trademark, but the next training to the product classification database may adjust some of the weight values back. Such back and forth adjustments may cause the weight values to fail to reach a good value. Training by database is equivalent to focusing on the weight values of the trademark classification database in one round of training, and then focusing on the weight values of the product title corresponding to the product classification database, and finally focusing on the weight values corresponding to the product classification database. This can better match the trademark and product classification. Judging from the training data, the training effect can improve the accuracy of about 3% compared with mixed training.
[0034] The main purpose of the above-mentioned database marking and training of massive title data is to distinguish between trademarks and product titles. For trademarks, they have a product classification, such as the Nice classification, but this classification belongs to the classification of trademarks. In the field of e-commerce, various e-commerce websites have not implemented the classification of products in such detail. Take the two product titles "Shandong Apple" and "Apple 15th Generation" as examples. The apple in "Shandong Apple" is a fruit, and the apple in "Apple 15th Generation" is a trademark. This method requires model training to first make "Shandong Apple" and "Apple 15th Generation" not in the same classification data, so that the trademark "apple" will not be confused with the "apple" in the fruit. For this reason, the technical solution of this application uses three databases to achieve the purpose of corresponding distinction.
[0035] Step S203, obtaining product title data to be processed, performing keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and inputting the input data into the trained model to obtain the trademark filtering and classification results of the input data.
[0036] The above-mentioned keyword extraction, key sentence extraction and data enhancement processing can all adopt existing methods, such as keyword extraction and key sentence extraction through natural language algorithms. Of course, other algorithms can also be used to perform the above-mentioned keyword extraction, key sentence extraction and data enhancement processing. The technical solution of this application is mainly aimed at targeted optimization of model training.
[0037] The technical solution provided by the present application collects and marks massive product title data, classifies the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database; inputs the massive title data as training data into the model to be trained to perform multiple rounds of iterative training to obtain a trained model, and a round of iterative training specifically includes: inputting the marked data in the trademark-to-trademark classification database as the first training data group into the model to be trained to perform the first round of training to obtain the first training model, inputting the marked data in the product title-to-product classification database as the second training data group into the first training model to perform the second round of training to obtain the second training model, and inputting the product classification database as the third training data group into the second training model to obtain the third training model to complete a round of iterative training. Obtain the product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain the trademark filtering classification result of the input data. The above scheme improves the training effect by performing step-by-step training on three types of databases, thereby improving the accuracy of recognition.
[0038] In an optional technical scenario, the above method may further include between step S201 and step S202:
[0039] Extract the number of layers n of the original training model, extract part of the data of a set ratio (e.g., 1 / 10, 1 / 8, etc.) from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database, respectively, to obtain the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database; use the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database as initial training data; determine the number of layers x of the original training model open for training, where x < n / 2, and divide the original training model into a first original training model, a second original training model ... an xth original training model according to the number of layers open for training, where the nth layer of each original training model in the first original training model, the second original training model ... the xth original training model is an open layer, and the remaining open layers are randomly selected open layers, and the open layers are layers whose weight data participate in iterative update during training, relative to If it is a non-open layer, the weight data corresponding to the non-open layer does not participate in the iterative update during training; the initial training data are respectively input into the first original training model, the second original training model...the xth original training model to perform training respectively to obtain the first training model, the second training model...the xth training model after training, obtain the first confidence rate, the second confidence rate...the xth confidence rate corresponding to the first training model, the second training model...the xth training model, extract the i-th original model corresponding to the maximum value of the first confidence rate, the second confidence rate...the x-th confidence rate, take i as the number of released layers, open and combine different levels to obtain y original models, the number of open layers of each model in the y original models is i layers and the nth layers are all open layers, input the initial training data into the y original models respectively to obtain y training models, obtain y confidence rates corresponding to the y training models, extract the i'th original model corresponding to the maximum value of the y confidence rates, and determine the i'th original model as the model to be trained.
[0040] The above technical solution is to confirm the number of open layers of the training model and the specific layer to be opened. For the training model, it has multiple layers. For example, taking the Bert model as an example, it has 12 layers, that is, n=12. However, in actual training, if all 12 layers are open layers, it is easy to overfit during training. Therefore, it is necessary to find a suitable open layer. Therefore, firstly determine the maximum number of open layers x based on experience. Taking 12 layers as an example, x=4. When taking x, generally x needs to be less than n / 2. In this way, first randomly select the original model with open layers of 1, 2...x to perform training, and then determine the maximum confidence rate of the trained model. In this way, the specific number of open layers i can be preliminarily determined, and then, when the number of open layers i remains unchanged, adjust the positions of i open layers to obtain y original models, and then select the original model with the highest confidence rate from the y original models as the model to be trained.
[0041] Taking i=2, n=12 as an example, the above y original models can be specifically 11 models, namely: open layer 1, 12-layer original model, open layer 2, 12-layer original model, open layer 3, 12-layer original model, open layer 4, 12-layer original model, open layer 5, 12-layer original model, open layer 6, 12-layer original model, open layer 7, 12-layer original model, open layer 8, 12-layer original model, open layer 9, 12-layer original model, open layer 10, 12-layer original model, open layer 11, 12-layer original model. Because the last layer is the output result layer, the last layer must be released for training. Therefore, when selecting an open layer, you only need to select the specific position of another open layer for verification, so as to achieve the training solution.
[0042] In an optional technical scenario, the obtaining of the first confidence rate, the second confidence rate, ... the xth confidence rate corresponding to the first training model, the second training model, ... the xth training model may specifically include:
[0043] Extract w pieces of data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database to obtain 3w pieces of sample data, and input the 3w pieces of sample data as input data to the first training model, the second training model, and the xth training model to obtain the first confidence rate, the second confidence rate, and the xth confidence rate, specifically including:
[0044] The 3w sample data are used as input data, and the first training model is used to perform calculations to obtain 3w confidence rates. The first confidence rate is determined to be the average of the 3w confidence rates, and the remaining x-1 training models are traversed to obtain the second confidence rate...the xth confidence rate.
[0045] In an optional technical scenario, the above-mentioned first training model using 3w sample data as input data to perform calculations to obtain 3w confidence rates may specifically include:
[0046] Extract the first sample data from the 3w sample data, input the first sample data as input data into the first training model to perform calculations to obtain multiple classification results and multiple confidence rates corresponding to the multiple classification results, query the first classification result corresponding to the marking result of the first sample data from the multiple classification results, extract the confidence rate corresponding to the first classification result and determine it as the confidence rate of the first sample data among the 3w confidence rates, and traverse the 3w sample data to obtain 3w confidence rates.
[0047] See also Figure 3 , Figure 3 Provide a structural block diagram of a system for obtaining trademark filtering and classification results through product titles, such as Figure 3 As shown, the above system includes:
[0048] The collection and classification unit 301 is used to collect and mark a large amount of product title data, and classify the large amount of product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database;
[0049] The training unit 302 is used to input the massive title data as training data into the model to be trained to perform multiple rounds of iterative training to obtain a trained model. One round of iterative training specifically includes: inputting the marked data in the trademark classification database of trademark pairs into the model to be trained as the first training data group to perform the first round of training to obtain the first training model, inputting the marked data in the product title pair of product classification database as the second training data group into the first training model to perform the second round of training to obtain the second training model, and inputting the product classification database as the third training data group into the second training model to obtain the third training model to complete one round of iterative training;
[0050] The execution unit 303 is used to obtain product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain the trademark filtering and classification results of the input data.
[0051] In an optional technical scenario, the system further includes:
[0052] The processing unit 304 is used to extract the number of layers n of the original training model, extract a set proportion of partial data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database, respectively, to obtain the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database; use the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database as initial training data; determine the number of layers x of the original training model for open training, where x < n / 2, and divide the original training model into a first original training model, a second original training model ... an xth original training model according to the number of layers for open training, where the nth layer of each original training model in the first original training model, the second original training model ... the xth original training model is an open layer, and the remaining open layers are randomly selected open layers, and the open layers are weighted during training. The value data participates in the iterative update of the layer; the initial training data are respectively input into the first original training model, the second original training model...the xth original training model to perform training respectively to obtain the first training model, the second training model...the xth training model after training, obtain the first confidence rate, the second confidence rate...the xth confidence rate corresponding to the first training model, the second training model...the xth training model, extract the i-th original model corresponding to the i-th confidence rate with the maximum value among the first confidence rate, the second confidence rate...the x-th confidence rate, take i as the number of released layers, open and combine different levels to obtain y original models, the number of open layers of each model in the y original models is i layers and the n-th layers are all open layers, input the initial training data into the y original models respectively to train to obtain y training models, obtain y confidence rates corresponding to the y training models, extract the i'th original model corresponding to the maximum value among the y confidence rates, and determine the i'th original model as the model to be trained.
[0053] In an optional technical scenario, the processing unit 304 is specifically used to extract w pieces of data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database to obtain 3w pieces of sample data, and input the 3w pieces of sample data as input data to the first training model, the second training model...the xth training model to obtain the first confidence rate, the second confidence rate...the xth confidence rate, specifically including:
[0054] The 3w sample data are used as input data, and the first training model is used to perform calculations to obtain 3w confidence rates. The first confidence rate is determined to be the average of the 3w confidence rates, and the remaining x-1 training models are traversed to obtain the second confidence rate...the xth confidence rate.
[0055] In an optional technical scenario, the processing unit 304 is specifically used to extract the first sample data from 3w sample data, input the first sample data as input data into the first training model to perform calculations to obtain multiple classification results and multiple confidence rates corresponding to the multiple classification results, query the first classification result corresponding to the marking result of the first sample data from the multiple classification results, extract the confidence rate corresponding to the first classification result and determine it as the confidence rate of the first sample data among the 3w confidence rates, and traverse the 3w sample data to obtain 3w confidence rates.
[0056] In an optional technical scenario, the above model is a Bert model.
[0057] The above processing unit can also be used to perform the following steps: Figure 2 The detailed or extended solutions of the illustrated method embodiment will not be described in detail here.
[0058] The embodiment of the present application can divide the electronic device into functional units according to the above method example. For example, each functional unit can be divided according to each function, or two or more functions can be integrated into one processing unit. The above integrated unit can be implemented in the form of hardware or in the form of software functional units. It should be noted that the division of units in the embodiment of the present application is schematic and is only a logical function division. There may be other division methods in actual implementation.
[0059] The above embodiments may be implemented in whole or in part by software, hardware, firmware or any other combination thereof. When implemented using software, the above embodiments may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions or computer programs. When the computer instructions or computer programs are loaded or executed on a computer, the process or function described in the embodiment of the present application is generated in whole or in part. The computer may be a general-purpose computer, a special-purpose computer, a computer network, or other programmable device. The computer instructions may be stored in a computer-readable storage medium, or transmitted from one computer-readable storage medium to another computer-readable storage medium. For example, the computer instructions may be transmitted from one website, computer, server or data center to another website, computer, server or data center by wired or wireless means. The computer-readable storage medium may be any available medium that a computer can access or a data storage device such as a server or data center that contains one or more available media sets. The available medium may be a magnetic medium (e.g., a floppy disk, a hard disk, a tape), an optical medium (e.g., a DVD), or a semiconductor medium. The semiconductor medium may be a solid-state hard disk.
[0060] An embodiment of the present application also provides a computer storage medium, wherein the computer storage medium stores a computer program for electronic data exchange, wherein the computer program enables a computer to execute part or all of the steps of any method described in the above method embodiments, and the above computer includes an electronic device.
[0061] The embodiment of the present application also provides a computer program product, the computer program product includes a non-transitory computer-readable storage medium storing a computer program, the computer program is operable to cause a computer to execute some or all of the steps of any method described in the method embodiment. The computer program product may be a software installation package, and the computer includes an electronic device.
[0062] It should be understood that in the various embodiments of the present application, the size of the serial numbers of the above-mentioned processes does not mean the order of execution. The execution order of each process should be determined by its function and internal logic, and should not constitute any limitation on the implementation process of the embodiments of the present application.
[0063] In the several embodiments provided in the present application, it should be understood that the disclosed methods, devices and systems can be implemented in other ways. For example, the device embodiments described above are merely schematic; for example, the division of the units is only a logical function division, and there may be other division methods in actual implementation; for example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.
[0064] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
[0065] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may be physically included separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of hardware plus software functional units.
[0066] The above-mentioned integrated unit implemented in the form of a software functional unit can be stored in a computer-readable storage medium. The above-mentioned software functional unit is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform some steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: a USB flash drive, a mobile hard disk, a magnetic disk, an optical disk, a volatile memory or a non-volatile memory. Among them, the non-volatile memory can be a read-only memory (ROM), a programmable read-only memory (PROM), an erasable programmable read-only memory (EPROM), an electrically erasable programmable read-only memory (EEPROM) or a flash memory. The volatile memory can be a random access memory (RAM), which is used as an external cache. By way of example but not limitation, many forms of random access memory (RAM) are available, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM) and direct RAM bus RAM (DR RAM). Various media that can store program code are available.
[0067] Although the present invention is disclosed as above, the present invention is not limited thereto. Any person skilled in the art can easily think of changes or substitutions without departing from the spirit and scope of the present invention, and can make various changes and modifications, including the combination of the above-mentioned different functions and implementation steps, including software and hardware implementation methods, all of which are within the scope of protection of the present invention.
Claims
1. A method for obtaining trademark filtering classification results through product titles, characterized in that: The method comprises the following steps: Collect and label massive product title data, and classify the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database; Inputting a large amount of title data as training data into the model to be trained and performing multiple rounds of iterative training to obtain a trained model. One round of iterative training specifically includes: inputting the marked data in the trademark classification database of trademark pairs into the model to be trained as the first training data group and performing the first round of training to obtain a first training model; inputting the marked data in the product classification database of product titles into the first training model as the second training data group and performing the second round of training to obtain a second training model; inputting the product classification database as the third training data group into the second training model and obtaining the third training model to complete one round of iterative training; Acquire product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain the trademark filtering and classification results of the input data; before inputting the massive title data as training data into the model to be trained to perform multiple rounds of iterative training to obtain the trained model, the method further includes: Extract the number of layers n of the original training model, extract a set proportion of partial data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database, respectively, to obtain the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database; use the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database as initial training data; determine the number of layers x of the original training model open for training, where x < n / 2, and divide the original training model into a first original training model, a second original training model ... an xth original training model according to the number of layers open for training, where the nth layer of each original training model in the first original training model, the second original training model ... the xth original training model is an open layer, and the remaining open layers are randomly selected open layers, and the weight data participates in the training of the open layers. Iteratively update the layer; input the initial training data into the first original training model, the second original training model...the xth original training model respectively, perform training on the trained first training model, the second training model...the xth training model respectively, obtain the first confidence rate, the second confidence rate...the xth confidence rate corresponding to the first training model, the second training model...the xth training model, extract the i-th original model corresponding to the i-th confidence rate with the maximum value among the first confidence rate, the second confidence rate...the x-th confidence rate, take i as the number of released layers, open and combine different levels to obtain y original models, the number of open layers of each model in the y original models is i layers and the n-th layers are all open layers, input the initial training data into the y original models respectively to train to obtain y training models, obtain y confidence rates corresponding to the y training models, extract the i'th original model corresponding to the maximum value among the y confidence rates, and determine the i'th original model as the model to be trained.
2. The method for obtaining trademark filtering and classification results through product titles according to claim 1, characterized in that: The step of obtaining the first confidence rate, the second confidence rate, ... the xth confidence rate corresponding to the first training model, the second training model, ... the xth training model specifically includes: Extract w pieces of data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database to obtain 3w pieces of sample data, and input the 3w pieces of sample data as input data to the first training model, the second training model, and the xth training model to obtain the first confidence rate, the second confidence rate, and the xth confidence rate, specifically including: The 3w sample data are used as input data, and the first training model is used to perform calculations to obtain 3w confidence rates. The first confidence rate is determined to be the average of the 3w confidence rates, and the remaining x-1 training models are traversed to obtain the second confidence rate...the xth confidence rate.
3. The method for obtaining trademark filtering and classification results through product titles according to claim 2, characterized in that: The first training model using 3w sample data as input data performs calculations to obtain 3w confidence rates specifically includes: Extract the first sample data from the 3w sample data, input the first sample data as input data into the first training model to perform calculations to obtain multiple classification results and multiple confidence rates corresponding to the multiple classification results, query the first classification result corresponding to the marking result of the first sample data from the multiple classification results, extract the confidence rate corresponding to the first classification result and determine it as the confidence rate of the first sample data among the 3w confidence rates, and traverse the 3w sample data to obtain 3w confidence rates.
4. The method for obtaining trademark filtering and classification results through product titles according to any one of claims 1 to 3, characterized in that: The model is the Bert model.
5. A system for obtaining trademark filtering and classification results through product titles, characterized in that: The system comprises: A collection and classification unit, used to collect and mark massive product title data, and classify the massive product title data to obtain a trademark-to-trademark classification database, a product title-to-product classification database, and a trademark and product classification database; A training unit is used to input a large amount of title data as training data into a model to be trained to perform multiple rounds of iterative training to obtain a trained model, wherein one round of iterative training specifically includes: inputting the marked data in the trademark classification database of trademark pairs into the model to be trained as a first training data group to perform a first round of training to obtain a first training model, inputting the marked data in the product classification database of product titles into the first training model as a second training data group to perform a second round of training to obtain a second training model, and inputting the product classification database as a third training data group into the second training model to obtain a third training model to complete one round of iterative training; An execution unit is used to obtain product title data to be processed, perform keyword extraction, key sentence extraction and data enhancement processing on the product title data to obtain input data, and input the input data into the trained model to obtain a trademark filtering and classification result of the input data; the system also includes: A processing unit is used to extract the number of layers n of the original training model, extract a set proportion of partial data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database, respectively, to obtain the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database; use the trademark-to-trademark classification partial database, the product title-to-product classification partial database, and the trademark and product classification partial database as initial training data; determine the number of layers x of the original training model open for training, where x < n / 2, and divide the original training model into a first original training model, a second original training model ... an xth original training model according to the number of layers open for training, where the nth layer of each original training model in the first original training model, the second original training model ... the xth original training model is an open layer, and the remaining open layers are randomly selected open layers, and the weights of the open layers are used during training. The data participates in the iterative update of the layer; the initial training data are respectively input into the first original training model, the second original training model...the xth original training model to perform training respectively to obtain the first training model, the second training model...the xth training model after training, obtain the first confidence rate, the second confidence rate...the xth confidence rate corresponding to the first training model, the second training model...the xth training model, extract the i-th original model corresponding to the i-th confidence rate with the maximum value among the first confidence rate, the second confidence rate...the x-th confidence rate, take i as the number of released layers, open and combine different levels to obtain y original models, the number of open layers of each model in the y original models is i layers and the n-th layers are all open layers, input the initial training data into the y original models respectively to train to obtain y training models, obtain y confidence rates corresponding to the y training models, extract the i'th original model corresponding to the maximum value among the y confidence rates, and determine the i'th original model as the model to be trained.
6. The system for obtaining trademark filtering and classification results by product title according to claim 5, characterized in that: The processing unit is specifically used to extract w pieces of data from the trademark-to-trademark classification database, the product title-to-product classification database, and the trademark and product classification database to obtain 3w pieces of sample data, and input the 3w pieces of sample data as input data to the first training model, the second training model...the xth training model to obtain the first confidence rate, the second confidence rate...the xth confidence rate, specifically including: The 3w sample data are used as input data, and the first training model is used to perform calculations to obtain 3w confidence rates. The first confidence rate is determined to be the average of the 3w confidence rates, and the remaining x-1 training models are traversed to obtain the second confidence rate...the xth confidence rate.
7. The system for obtaining trademark filtering and classification results by product title according to claim 6, characterized in that: The processing unit is specifically used to extract the first sample data from 3w pieces of sample data, input the first sample data as input data into the first training model to perform calculations to obtain multiple classification results and multiple confidence rates corresponding to the multiple classification results, query the first classification result corresponding to the marking result of the first sample data from the multiple classification results, extract the confidence rate corresponding to the first classification result and determine it as the confidence rate of the first sample data from the 3w confidence rates, and traverse the 3w pieces of sample data to obtain 3w confidence rates.
8. The system for obtaining trademark filtering and classification results by product title according to any one of claims 5 to 7, characterized in that: The model is the Bert model.
