Method and device for establishing label matching detection model, storage medium and equipment

Abstract

The present disclosure provides a method and device for establishing a label matching degree detection model, a storage medium and an electronic device; it relates to the field of machine learning. The method includes: obtaining user feature data and user operation data, wherein the user operation data includes search data and selection operation; obtaining label data based on the search data, and obtaining the corresponding relationship between the item and the label data based on the selection operation, wherein the item is the item searched by the search operation corresponding to the search data; determining the positive sample label data of the item based on the corresponding relationship, and obtaining the negative sample label data of the item; obtaining training data based on the positive and negative sample label data of the item, item information, the user operation data and the user feature data, and based on the training data, using a machine learning algorithm to train to obtain the label matching degree detection model. The present disclosure can detect the matching degree between labels and items.

Description

Technical Field

[0001] The present disclosure relates to the field of machine learning, and more specifically, to a method for establishing a label matching detection model, a device for establishing a label matching detection model, a label matching detection method, a label matching detection device, an electronic device, and a computer-readable storage medium. Background Art

[0002] In many cases, adding tags to items can facilitate operations such as statistics, search, and transactions, and the degree of match between items and tags has an important impact on these operations. Taking e-commerce platforms as an example, merchants can increase the exposure and search hit rate of products by adding tags to products. However, only when the degree of match between products and their tags is high, the products recalled based on user search operations will be more in line with user needs and increase the user's click-through rate for products.

[0003] In actual situations, the matching degree between objects and labels will be affected by various factors such as human factors. Therefore, some means need to be taken to detect the matching degree, but the existing methods such as manual review or text similarity judgment have problems such as low efficiency, strong subjectivity, low operability and low accuracy.

[0004] Therefore, it is necessary to propose a method for establishing a label matching detection model. The model established by this method can more accurately determine the matching degree between objects and labels, improve the speed of matching detection, improve the efficiency of data processing, and save computer and human resources.

[0005] The information disclosed in the above background section is only for enhancing the understanding of the background of the present disclosure and therefore may contain information that does not constitute the prior art known to a person of ordinary skill in the art. Summary of the invention

[0006] The embodiments of the present disclosure provide a method and device for establishing a label matching detection model, a method and device for establishing a label matching detection model, an electronic device, and a computer-readable storage medium. The model established by the method for establishing a label matching detection model can more accurately determine the matching degree between an object and a label, thereby increasing the speed of matching detection, improving the efficiency of data processing, and saving computer and human resources.

[0007] According to a first aspect of the present disclosure, a method for establishing a label matching detection model is provided, comprising:

[0008] Acquire user feature data and user operation data, wherein the user operation data includes search data and selection operations;

[0009] Acquire label data based on the search data, and acquire a corresponding relationship between an item and the label data based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data;

[0010] Determine the positive sample label data of the object according to the corresponding relationship, and obtain the negative sample label data of the object;

[0011] The training data is obtained according to the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and based on the training data, the label matching degree detection model is obtained by training using a machine learning algorithm.

[0012] In an exemplary embodiment of the present disclosure, the acquiring tag data based on the search data includes:

[0013] Obtain a list of labels in a label library for classifying the items;

[0014] The search data is text-matched with the tags in the tag list, and tag data corresponding to the search data is acquired based on the result of the text matching.

[0015] In an exemplary embodiment of the present disclosure, the acquiring the corresponding relationship between the item and the tag data based on the selection operation includes:

[0016] Whether an item is selected is determined based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data.

[0017] In an exemplary embodiment of the present disclosure, determining the positive sample label data of the object according to the corresponding relationship includes:

[0018] When the object is selected, the label data is used as positive sample label data of the object.

[0019] In an exemplary embodiment of the present disclosure, the obtaining of negative sample label data of the object includes:

[0020] When the object is selected, a plurality of labels other than the label data are extracted from the label library as negative sample label data of the object.

[0021] In an exemplary embodiment of the present disclosure, the step of obtaining training data based on the positive and negative sample label data of the item, the item information, the user operation data, and the user feature data includes:

[0022] Preprocessing the positive and negative sample label data of the item, item information, the user operation data and the user feature data, wherein the preprocessing includes data deduplication, missing value completion and screening;

[0023] The preprocessed positive and negative sample label data of the item, item information, the user operation data and the user feature data are vectorized to obtain training data.

[0024] According to a second aspect of the present disclosure, a tag matching degree detection method is provided, comprising:

[0025] Establish a label matching detection model according to the above method;

[0026] The data to be tested is obtained and input into the label matching detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

[0027] In an exemplary embodiment of the present disclosure, the inputting into the label matching detection model to obtain the matching degree between the object and the label includes:

[0028] Inputting the test data into the label matching detection model to obtain a probability score of the object being selected under the label corresponding to the test data;

[0029] The matching degree between the object and the label is calculated according to the probability score.

[0030] According to a third aspect of the present disclosure, a device for establishing a label matching degree detection model is provided, comprising:

[0031] A data acquisition module, used to acquire user feature data and user operation data, wherein the user operation data includes search data and selection operations;

[0032] a relationship determination module, configured to obtain label data based on the search data, and obtain a corresponding relationship between an item and the label data based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data;

[0033] A data processing module, used to determine the positive sample label data of the object according to the corresponding relationship, and obtain the negative sample label data of the object;

[0034] The model training module is used to obtain training data based on the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and based on the training data, use a machine learning algorithm to train the label matching detection model.

[0035] According to a fourth aspect of the present disclosure, a label matching degree detection device is provided, comprising:

[0036] A model building module, used to build a label matching detection model according to the above method;

[0037] The matching degree detection module is used to obtain the data to be tested and input it into the label matching degree detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

[0038] According to a fifth aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored, and when the computer program is executed by a processor, any one of the above methods is implemented.

[0039] According to a sixth aspect of the present disclosure, there is provided an electronic device, including:

[0040] processor;

[0041] A memory for storing executable instructions of the processor; wherein the processor is configured to perform any one of the above methods by executing the executable instructions.

[0042] The exemplary embodiments of the present disclosure may have some or all of the following beneficial effects:

[0043] In the method for establishing a label matching degree detection model provided in the exemplary implementation of the present disclosure, first, user feature data and user operation data are obtained, wherein the user operation data includes search data and selection operation; then, label data is obtained based on the above search data, and the corresponding relationship between the item and the label data is obtained based on the above selection operation, wherein the item is the item searched by the search operation corresponding to the above search data; after obtaining the above corresponding relationship, the positive sample label data of the item is determined according to the corresponding relationship, and the negative sample label data of the item is obtained; finally, training data is obtained according to the positive and negative sample label data of the item, item information, user operation data and user feature data, and a label matching degree detection model is obtained by training using a machine learning algorithm based on the training data. On the one hand, the exemplary implementation of the present disclosure uses feature data of two dimensions of positive and negative sample label data when training the model, and also uses multiple data such as item information, user feature data and user operation data, which can improve the accuracy of the established model, better judge the matching degree between the item and its label, and improve the accuracy of the item operation. On the other hand, the model established by the method provided by the exemplary embodiment of the present disclosure can detect the matching degree of objects and object labels without manual review, thus saving labor costs. At the same time, it also avoids the deviation caused by human subjective cognition.

[0044] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0045] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0046] Figure 1 A schematic diagram showing an exemplary system architecture of a method and device for establishing a label matching degree detection model to which the embodiments of the present disclosure can be applied;

[0047] Figure 2 A schematic diagram showing the structure of a computer system suitable for implementing an electronic device of an embodiment of the present disclosure is shown;

[0048] Figure 3 A flowchart schematically illustrates a process of establishing a tag matching degree detection model according to an embodiment of the present disclosure;

[0049] Figure 4A flowchart schematically shows a process of a tag matching degree detection method according to an embodiment of the present disclosure;

[0050] Figure 5 A block diagram schematically shows a device for establishing a label matching degree detection model according to an embodiment of the present disclosure;

[0051] Figure 6 A block diagram of a label matching degree detection device according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0052] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as being limited to the examples set forth herein; on the contrary, these embodiments are provided so that the present disclosure will be more comprehensive and complete, and the concepts of the example embodiments are fully conveyed to those skilled in the art. The described features, structures, or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present disclosure. However, those skilled in the art will appreciate that the technical solutions of the present disclosure may be practiced while omitting one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known technical solutions are not shown or described in detail to avoid obscuring various aspects of the present disclosure.

[0053] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0054] Figure 1 A schematic diagram of a system architecture of an exemplary application environment in which a method and device for establishing a label matching degree detection model according to an embodiment of the present disclosure can be applied is shown.

[0055] like Figure 1As shown, the system architecture 100 may include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links or optical fiber cables, etc. The terminal devices 101, 102, 103 may be various electronic devices with display screens, including but not limited to desktop computers, portable computers, smart phones, tablet computers, etc. It should be understood that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration. According to the implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

[0056] The method for establishing the label matching degree detection model provided in the embodiment of the present disclosure can be executed by the terminal devices 101, 102, and 103, and accordingly, the device for establishing the label matching degree detection model can be set in the terminal devices 101, 102, and 103. The method for establishing the label matching degree detection model provided in the embodiment of the present disclosure can also be executed by the server 105, and accordingly, the device for establishing the label matching degree detection model can be set in the server 105. The method for establishing the label matching degree detection model provided in the embodiment of the present disclosure can also be jointly executed by the terminal devices 101, 102, and 103 and the server 105, and accordingly, the device for establishing the label matching degree detection model can be set in the terminal devices 101, 102, and 103 and the server 105, which is not particularly limited in this exemplary embodiment.

[0057] For example, in this example implementation, user feature data and user operation data can be obtained through terminal devices 101, 102, and 103, wherein the user operation data includes search data and selection operations; then, terminal devices 101, 102, and 103 can send the obtained data to server 105 through network 104, and server 105 obtains label data based on the search data, and obtains the corresponding relationship between the item and the label data based on the selection operation, wherein the item is the item searched for by the search operation corresponding to the search data, and the positive sample label data of the item is determined based on the corresponding relationship, and the negative sample label data of the item is obtained; finally, server 1 obtains training data based on the positive and negative sample label data of the item, item information, user operation data, and user feature data, and based on the obtained training data, uses a machine learning algorithm to train a label matching detection model.

[0058] Figure 2 A schematic diagram of the structure of a computer system suitable for implementing an electronic device of an embodiment of the present disclosure is shown.

[0059] It should be noted that Figure 2 The computer system 200 of the electronic device shown is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0060] like Figure 2 As shown, the computer system 200 includes a central processing unit (CPU) 201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 202 or a program loaded from a storage part 208 into a random access memory (RAM) 203. Various programs and data required for system operation are also stored in the RAM 203. The CPU 201, the ROM 202, and the RAM 203 are connected to each other via a bus 204. An input / output (I / O) interface 205 is also connected to the bus 204.

[0061] The following components are connected to the I / O interface 205: an input section 206 including a keyboard, a mouse, etc.; an output section 207 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 208 including a hard disk, etc.; and a communication section 209 including a network interface card such as a LAN card, a modem, etc. The communication section 209 performs communication processing via a network such as the Internet. A drive 210 is also connected to the I / O interface 205 as needed. A removable medium 211, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 210 as needed, so that a computer program read therefrom is installed into the storage section 208 as needed.

[0062] In many cases, adding tags to items can facilitate operations such as statistics, search, and transactions, and the degree of match between items and tags has an important impact on these operations. Taking e-commerce platforms as an example, merchants can increase the exposure and search hit rate of products by adding tags to products. However, only when the degree of match between products and their tags is high, the products recalled based on user search operations will be more in line with user needs and improve the selection conversion rate.

[0063] In actual situations, the matching degree between the object and the tag may be affected by various factors such as human factors. Therefore, some means are needed to detect the matching degree. In order to achieve the goal of this detection, the inventor tentatively proposed the following two methods:

[0064] The first method is manual review, that is, the reviewer reviews the labels of the items based on his or her understanding of the items and the labels in the label library. Taking the e-commerce platform as an example, this method is implemented as follows: the merchant's backend operation personnel select the most appropriate marketing label based on their understanding of the product and the marketing label system. For example, for a product titled "Brand A Casual Pants Men 2019 Autumn and Winter Thick Slim Straight Slim-fit Men's Pants Korean Youth Business Suit Pants Elastic Retro Long Pants Men 1047 Khaki 31", the merchant may give the product marketing labels including "Korean Style", "Business Casual", "Retro Style", "Motorcycle Style", etc. The reviewer reviews these labels based on his or her own experience, and approves high-quality labels with high matching degree, and rejects low-quality labels. In this method, if the reviewer is experienced, the accuracy of the reviewed item labels is relatively high. However, this method also has the following problems: (1) The speed and quality of the review by the reviewer will be affected and limited by the reviewer's energy, requiring a lot of manual labor; (2) In most cases, the item labels will be dynamically updated, and it is very difficult for the reviewer to remember the effective values ​​of the marketing label library in real time and make accurate judgments.

[0065] The second method is to judge the matching degree based on algorithms such as short text semantic similarity judgment and online short text clustering judgment. This method first converts the short text into a semantic vector representation, and then calculates the similarity of the two short texts. However, the main problem of this method is the complexity of text semantics, and the judgment result is not accurate. For example, the short text semantic similarity judgment method is mainly based on the statistical implementation of the characters or words in the text, which is greatly affected by the inherent noise of the text. If some irrelevant words appear repeatedly in the text, it will lead to huge differences in the text vector representation, and then lead to completely different calculation results; and the method based on online short text clustering, in the clustering process, because the amount of online text is difficult to control, and the similarity threshold needs to change dynamically with the number of aggregated short texts, its calculation results often have great uncertainty and inoperability.

[0066] In order to solve the problems existing in the above method, in this exemplary embodiment, the inventor proposes a new technical solution. The technical solution of the embodiment of the present disclosure is described in detail below:

[0067] This exemplary embodiment first provides a method for establishing a label matching degree detection model. The method for establishing a label matching degree detection model is as follows: Figure 3 As shown, the specific steps include:

[0068] Step S310: Acquire user feature data and user operation data, wherein the user operation data includes search data and selection operations;

[0069] Step S320: acquiring label data based on the search data, and acquiring a corresponding relationship between an item and the label data based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data;

[0070] Step S330: determining the positive sample label data of the object according to the corresponding relationship, and obtaining the negative sample label data of the object;

[0071] Step S340: obtaining training data according to the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and obtaining the label matching degree detection model by training using a machine learning algorithm based on the training data.

[0072] In the method for establishing a label matching degree detection model provided in the disclosed example implementation, on the one hand, the disclosed example implementation adopts feature data of two dimensions, namely, positive and negative sample label data, when training the model, and also utilizes multiple data such as item information, user feature data, and user operation data, which can improve the accuracy of the established model, better judge the matching degree between items and their labels, and improve the accuracy of item operations. On the other hand, the model established by the method provided in the disclosed example implementation can detect the matching degree between items and item labels without manual review, thus saving labor costs. At the same time, it also avoids the deviation caused by subjective cognition of people in judgment.

[0073] Next, in another embodiment, the above steps are described in more detail.

[0074] In step S310, user characteristic data and user operation data are obtained, wherein the user operation data includes search data and selection operations.

[0075] In this example implementation, user characteristic data is data used to describe user characteristics, and may be user attribute data, electronic transaction data of users on e-commerce platforms, or other data that can characterize user characteristics, or a combination of the above-mentioned data. Taking the e-commerce platform as an example, the user characteristics may include data such as the user's age, gender, purchasing power level, and city where they are located. This example implementation does not make any special limitations for comparison.

[0076] In this example implementation, user operation data is a series of operations performed by the user to achieve a certain purpose. The operation data may include search data and selection operations, and may also include other operations such as viewing, deleting, etc. This example implementation does not specifically limit this.

[0077] Among them, the above-mentioned search data are keywords used by users to search for supplies, and the selection operation is a selection operation performed by the user to select the items he needs from the items searched by the search operation corresponding to the search data. The selection operation can be a click operation, a re-press operation, or other operations that meet the above definition. This example implementation does not make any special limitations on this.

[0078] The above definition is explained by taking the e-commerce platform as an example: in order to achieve the purpose of purchasing a certain product, the user can perform a series of corresponding operations to generate user operation data. For example, according to one's own purchasing needs, the keywords used by the user in the e-commerce platform client can be understood as search data, and based on the user's search operation, the e-commerce platform will recall some products to the client, and the user chooses to view the products that meet his needs, which can be understood as a selection operation. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0079] The method for establishing the tag matching degree detection model provided in this example implementation can obtain the above-mentioned user feature data and user operation data through the terminal device. Taking the e-commerce platform as an example, the user feature data and user operation data required by the method can be obtained through the application client of the e-commerce platform in the terminal device. In addition, the above-mentioned data can also be obtained through other channels and methods, and this example implementation does not specifically limit this.

[0080] In step S320, label data is acquired based on the search data, and a corresponding relationship between an item and the label data is acquired based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data.

[0081] In this example implementation, the item is the item obtained by the search operation corresponding to the above search data, the tag data is the tag related to the item in the search data, and the above item is obtained by searching through the tag. Taking the e-commerce platform as an example, the item can be the commodity searched by the user, the search data can be the keyword input by the user to describe the relevant characteristics of the desired commodity, and the tag data can be the commodity tag extracted from the search data. For example, if a user wants to buy a jacket with a small fresh style, the search data can be "small fresh jacket", where "small fresh" can be understood as the tag data extracted from the above search data. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0082] In this example implementation, the process of obtaining label data based on search data can be implemented as follows, for example: obtaining a label list in a label library for classifying the above-mentioned items; performing text matching between the above-mentioned search data and the labels in the label list, and obtaining label data corresponding to the search data based on the result of the text matching. The text matching method can be a dictionary tree, a short text similarity calculation, or other methods that can achieve text matching, which is not specifically limited in this example implementation.

[0083] The following takes the e-commerce platform as an example to further explain the process of obtaining label data based on search data:

[0084] First, obtain the currently valid tag list from the tag library of the e-commerce platform. For example, the tag list can be ['Hong Kong style', 'British style', 'Youth leisure', 'Business formal', 'Business leisure', 'European and American style', 'Retro style', 'Chinese style', 'Military workwear', 'Korean version', 'Street', 'Punk', 'Hip-hop', 'Japanese style', 'Original design', 'Basic mass', 'Japanese and Korean style', 'Original ecology', 'Small fresh', 'Casual style', 'Neutral style'..., 'Sports leisure']; then, perform text matching between the search data entered by the user and the tags in the tag list, and extract the tag with the highest text matching degree as the tag data corresponding to the search data. For example, if the search keyword entered by the user is "small fresh jacket", the tag with the highest text matching degree with the search data in the tag list is "small fresh", so small fresh is the tag data corresponding to the user's search data. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0085] In this example implementation, the user's search operation can obtain multiple items that meet the keyword characteristics of his search data, and the user can select items that meet his requirements. The user's selection operation can be a click operation, a long press operation, or other operations that can achieve the same function. This example implementation does not specifically limit this.

[0086] In this example implementation, based on the user's selection operation, the corresponding relationship between the searched item and the tag data can be obtained. Taking the selection operation as a click operation as an example, the corresponding relationship can be specifically reflected in whether the searched item is clicked by the user under the tag data, and the timestamp of the click, etc. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0087] In step S330, the positive sample label data of the object is determined according to the corresponding relationship, and the negative sample label data of the object is obtained.

[0088] In this example implementation, after the correspondence between the object and the label data is obtained in step S320, the positive sample data of the object can be determined based on the correspondence. The determination process can be: when the object is clicked, the label data is used as the positive sample label data of the object.

[0089] Taking the e-commerce platform as an example, the positive sample data of the above-mentioned determined items are explained as follows: Assuming that the search data entered by the user is "small fresh coat", the label data obtained based on the search data is "small fresh", and the user's search operation recalls a total of six items from product A to product F. If product A meets the user's requirements, the user will click on the product A to view the product details. At this time, the label data "small fresh" is the positive sample label data of product A. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0090] In this example implementation, while determining the positive sample data of the item, the negative sample data of the item can also be obtained based on the corresponding relationship. The process can be: when the item is clicked, multiple labels other than the label data are extracted from the label library as the negative sample label data of the item.

[0091] Among them, extracting multiple labels other than the label data as negative sample label data of the object can be completed by random sampling, that is, according to the actual needs of training, randomly sampling multiple labels other than the label data from the label list of the label library in proportion as the negative sample labels of the object.

[0092] Taking the above e-commerce platform as an example, the process is explained as follows: Assume that the above tag list is ['Hong Kong style', 'British style', 'Youth casual', 'Business formal', 'Business casual', 'European and American style', 'Retro style', 'Chinese style', 'Military workwear', 'Korean style', 'Street', 'Punk', 'Hip-hop', 'Japanese style', 'Original design', 'Basic mass', 'Japanese and Korean style', 'Original ecology', 'Small fresh', 'Casual style', 'Neutral style'..., 'Sports casual'], and the search data entered by the user is "Small fresh New coat", the label data obtained based on the search data is "little fresh", the user's search operation recalls a total of six products from product A to product F, if product A meets the user's requirements, the user will click on product A to view product details. At this time, labels other than "little fresh" in the label list, such as "business formal", "business casual", "European and American style", "retro style", etc. can all be used as negative sample label data for product A, and the negative sample label data of product A can be obtained in proportion by random sampling from the label list. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0093] In step S340, training data is obtained based on the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and based on the training data, a machine learning algorithm is used to train the label matching detection model.

[0094] In this example implementation, item information includes multiple pieces of information that can characterize item characteristics, such as name, title, category, etc. Taking the e-commerce platform as an example, the above item information is commodity information, which may include multiple pieces of information such as commodity title, color, category, whether it is self-operated, etc. It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0095] In this example implementation, the training data of the model can be obtained based on the positive and negative sample label data of the item obtained in step S330, as well as the above-mentioned item information, user operation data and user feature data. The process can be: preprocessing the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, wherein the preprocessing includes data deduplication, missing value completion and screening; vectorizing the preprocessed data to obtain the training data.

[0096] The following is a detailed description of the process of obtaining training data using an e-commerce platform as an example:

[0097] First, align the positive and negative sample data of the obtained products with product information, user operation data, and user characteristic data. Taking the list of valid tag values in the label library of an e-commerce platform as ['Hong Kong style', 'British style', 'Youthful casual', 'Business formal', 'Business casual', 'European and American style', 'Retro style', 'Chinese style', 'Military workwear', 'Korean style', 'Street style', 'Punk style', 'Hip-hop style', 'Japanese style', 'Original design', 'Basic and popular', 'Japanese and Korean style', 'Ecological', 'Fresh', 'Casual style', 'Neutral style'...,'Sports casual'] as an example, the data alignment process is illustrated through Table 1 and Table 2: In Table 1, based on the search data of user user_a, the tag matched from the above list of valid tag values is 'Fresh', and user user_a viewed the products with titles name1 and name2 recalled by the e-commerce platform through click operations. Align the above product information, user operation data, and user characteristic data according to fields such as product ID and account to obtain the search click alignment data shown in Table 1. In Table 2, based on the search data of user user_b, the tag matched from the above list of valid tag values is also 'Fresh'. The titles of the products recalled by the e-commerce platform are name3 and name4, but they were not clicked by the user. Align the above product information, user operation data, and user characteristic data according to fields such as product ID and account to obtain the search exposure (not clicked) data shown in Table 2.

[0098] Table 1

[0099]

[0100] Among them, the products with product IDs '123456' and '151256' were searched and clicked by user user_a at two time points: 2019-11-30 12:05 and 2019-11-30 12:07 respectively.

[0101] Table 2

[0102]

[0103] Among them, the products with product IDs '56788' and '56799' were searched by user user_b but not clicked.

[0104] Next, perform feature selection and preprocessing on the aligned data obtained above. First, important features such as product title, color, category, whether it is self-operated, user click time, search query, user age, gender, purchasing power level, and city can be manually selected according to experience. For example, boys around the age of 20 may like styles such as "hip-hop" and "motorcycle", so features such as age and gender are all important features. Then, perform data preprocessing tasks such as data deduplication and missing value filling on the data. If a data field is missing too severely, the column feature is removed.

[0105] In the above process, if products with titles name1 and name2 in the products are clicked by users, then "fresh" can be used as the positive sample label data for products name1 and name2. Then, several other labels except "fresh", such as "street", "punk", "hip-hop", etc., are randomly selected from the valid label value list in proportion as their negative sample label data to adjust the ratio of positive and negative samples. In some cases, the labels in the search keywords corresponding to the unclicked products can also be used as the negative sample label data of the products. As shown in Table 2, "fresh" can be used as the positive sample label for products name3 and name4. After determining the positive and negative sample label data of the products, perform vectorization processing on the preprocessed positive and negative sample label data, item information, user operation data, and user feature data. This vectorization processing can be achieved by the method of feature hashing and taking the remainder, and finally obtain the training data.

[0106] It should be noted that the above scenario is only an exemplary illustration, and the protection scope of the present exemplary embodiment is not limited thereto.

[0107] In the present exemplary embodiment, after obtaining the above training data, based on the training data, a label matching degree detection model can be trained using a machine learning algorithm. For example, the FTRL (Follow-the-regularized-Leader, an approximate online learning algorithm) loss function can be used to update the parameters during the training process. Its idea is to find the parameters that minimize the sum of the loss functions of all previous samples in each iteration. The loss function is as follows:

[0108]

[0109] The above formula consists of four terms. The first term is an estimate of the contribution to the loss function; the second and third terms represent the L1 and L2 regularization terms. Among them, the former is used to obtain a sparse solution, and the latter can make the solution result more "smooth"; the fourth term is to control that w does not change too much in each iteration, and σ s represents the set learning rate. Among them, W is the model parameter, and λ1 and λ2 are the L1 and L2 regularization coefficients respectively.

[0110] It should be noted that the above scenario is only an exemplary description, and the use of other algorithms for model training also falls within the protection scope of this example implementation.

[0111] After the label matching detection model is obtained through training in the above steps S310 to S340, this exemplary embodiment also provides a label matching detection method to detect the matching degree between the object label and the object through the above label matching detection method. Figure 4 As shown, it includes the following processes:

[0112] Step S410: Establishing a label matching degree detection model according to the method described in steps S310 to S340 above;

[0113] Step S420: Acquire the data to be tested and input it into the label matching detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

[0114] Next, in another embodiment, the above steps are described in more detail.

[0115] In step S410, a tag matching degree detection model is established according to the method described in steps S310 to S340. This establishment process has been described in detail above, so it will not be repeated here.

[0116] In step S420, the data to be tested is obtained and input into the label matching detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

[0117] In this example implementation, the data to be tested is data used to detect the matching degree between the item and the label added to the item, and may include item information, labels, user feature data, user operation data and other related data. The data to be tested may be data obtained by aligning the above data according to the item ID in the item information, or may be other forms of data that meet the above definition, and this example implementation does not specifically limit this.

[0118] Taking the e-commerce platform as an example, when the above-mentioned data to be tested is data obtained by aligning the above-mentioned data according to the item ID in the item information, the data to be tested may be as shown in Table 3:

[0119] Table 3

[0120]

[0121] The tags in Table 3 are tags added by merchants to products, which are different from those in Table 1 and Table 2, which are extracted based on the user's search data. The records in Table 3 are the data records of product 13579 being searched and clicked.

[0122] It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0123] In this example implementation, after the above-mentioned test data is obtained, the tag matching degree detection model established by the above-mentioned steps S310 to S340 can obtain the matching degree between the object and the tag in the above-mentioned test data. The implementation of this process can be as follows: input the test data into the tag matching degree detection model to obtain the probability score of the object being clicked under the tag corresponding to the test data; and calculate the matching degree between the object and the tag based on the probability score.

[0124] Below, taking the e-commerce platform as an example, the above process is explained in detail:

[0125] Input multiple prediction data as shown in Table 3 into the established label matching detection model, and make predictions based on each piece of data to be tested, and the prediction results shown in Table 4 can be obtained:

[0126] Table 4

[0127] Product ID Label marketing label Prediction score 1 13579 Small fresh 0.32 2 13579 Small fresh 0.28 3 13579 Korean version 0.06 4 13579 Korean version 0.04

[0128] After obtaining the prediction results shown in Table 4, due to the non-uniqueness of the product ID in the result data, the products are grouped according to the marketing tags, and the average matching degree of each group is calculated as the final score according to the formula shown below:

[0129]

[0130] Where k represents the number of data items of a certain product marketing label after grouping in the prediction result. i Indicates the matching value between each marketing attribute and product predicted by the model.

[0131] By further calculating the prediction results shown in Table 4 using the above formula, we can obtain the results shown in Table 5:

[0132] Table 5

[0133] Product ID Label marketing label score_avreage score_max_sku score_max_label 1 13579 Small fresh 0.30 0.30 0.35 2 13579 Korean version 0.05 0.30 0.12

[0134] Among them, from the product dimension (score_max_sku), for the product with ID 13579, the label with the highest score is "little fresh", with a score of 0.30, and the label with the lowest score is "Korean version", with a score of 0.05; from the label dimension (score_max_label), among all the product data with the "little fresh" marketing label, the highest score is 0.35.

[0135] Therefore, for a product with ID 13579, when it is labeled "little fresh", the probability of being clicked is 0.30, and it is most likely to be searched and clicked by users; when it is labeled "Korean version", the probability of being clicked is 0.05, and the possibility of being searched and clicked is very small; therefore, the model can conclude that the product has a higher match with the label "little fresh".

[0136] The model building and utilization method of this example implementation is applied to the e-commerce platform, and the matching degree between the product and the label is verified by combining the product dimension, user dimension, search click and other dimensions, avoiding the subjectivity of manual labeling. At the same time, this method improves the accuracy of search recommendations, reduces invalid traffic, and improves user experience.

[0137] It should be noted that the above scenario is only an exemplary description, and the protection scope of this example implementation is not limited to this.

[0138] It should be noted that the above application scenario is only an exemplary description. Although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0139] Furthermore, this exemplary embodiment provides a device for establishing a label matching degree detection model. Figure 5 As shown, the device 500 for establishing the label matching degree detection model may include a data acquisition module 510 , a relationship judgment module 520 , a data processing module 530 and a model training module 540 .

[0140] in:

[0141] The data acquisition module 510 may be used to acquire user characteristic data and user operation data, wherein the user operation data includes search data and click data;

[0142] The relationship determination module 520 may be used to obtain label data based on the search data, and to obtain a corresponding relationship between an item and the label data based on the click data, wherein the item is an item searched by the search operation corresponding to the search data;

[0143] The data processing module 530 may be used to determine the positive sample label data of the object according to the corresponding relationship, and obtain the negative sample label data of the object;

[0144] The model training module 540 can be used to obtain training data based on the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and based on the training data, use a machine learning algorithm to train the label matching detection model.

[0145] The specific details of each module or unit in the above-mentioned label matching degree detection model establishment device have been described in detail in the corresponding label matching degree detection model establishment method, so they will not be repeated here.

[0146] Correspondingly, with respect to the above-mentioned tag matching degree detection method, this exemplary embodiment also provides a tag matching degree detection device. Figure 6 As shown, the label matching degree detection device 600 may include a model building module 610 and a matching degree detection module 620. Among them:

[0147] The model building module 610 can be used to build a label matching detection model according to the method described in steps S310 to S340 above;

[0148] The matching detection module 620 can be used to obtain the data to be tested and input it into the label matching detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, the item information of the item and user operation data.

[0149] The specific details of each module or unit in the above-mentioned label matching degree detection device have been described in detail in the corresponding label matching degree detection method, so they will not be repeated here.

[0150] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0151] As another aspect, the present application also provides a computer-readable medium, which may be included in the electronic device described in the above embodiment; or may exist independently without being assembled into the electronic device. The above computer-readable medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the above embodiment. For example, the electronic device may implement Figure 3-4 The various steps shown, etc.

[0152] It should be noted that the computer-readable medium shown in the present disclosure may be a computer-readable signal medium or a computer-readable storage medium or any combination of the above two. The computer-readable storage medium may be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device or device, or any combination of the above. More specific examples of computer-readable storage media may include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in combination with an instruction execution system, device or device. In the present disclosure, a computer-readable signal medium may include a data signal propagated in a baseband or as part of a carrier wave, in which a computer-readable program code is carried. This propagated data signal may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. Computer-readable signal media may also be any computer-readable medium other than computer-readable storage media, which may send, propagate or transmit a program for use by or in conjunction with an instruction execution system, apparatus or device. The program code contained on the computer-readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wire, optical cable, RF, etc., or any suitable combination of the above.

[0153] It should be understood that the present disclosure is not limited to the exact structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A method for establishing a label matching detection model, characterized in that: include: Acquire user feature data and user operation data, wherein the user operation data includes search data and selection operations; Acquire tag data based on the search data, and determine whether an item is selected based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data; When the item is selected, the label data is used as the positive sample label data of the item; when the item is selected, multiple labels other than the label data are extracted from the label library as the negative sample label data of the item; training data is obtained based on the positive and negative sample label data of the item, item information, the user operation data and the user feature data, and based on the training data, the label matching detection model is trained using a machine learning algorithm.

2. The method for establishing a label matching detection model according to claim 1, characterized in that: The acquiring tag data based on the search data includes: Obtain a list of labels in a label library for classifying the items; The search data is text-matched with the tags in the tag list, and tag data corresponding to the search data is acquired based on the result of the text matching.

3. The method for establishing a label matching detection model according to claim 1, characterized in that: The obtaining of training data based on the positive and negative sample label data of the item, the item information, the user operation data and the user feature data includes: Preprocessing the positive and negative sample label data of the item, item information, the user operation data and the user feature data, wherein the preprocessing includes data deduplication, missing value completion and screening; The preprocessed positive and negative sample label data of the item, item information, the user operation data and the user feature data are vectorized to obtain training data.

4. A tag matching detection method, characterized in that: include: Establishing a label matching detection model according to the method described in any one of claims 1 to 3; The data to be tested is obtained and input into the label matching detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

5. The tag matching detection method according to claim 4, characterized in that: The input to the label matching detection model to obtain the matching degree between the object and the label in the test data includes: Inputting the test data into the label matching detection model to obtain a probability score of the object being selected under the label corresponding to the test data; The matching degree between the object and the label is calculated according to the probability score.

6. A device for establishing a label matching degree detection model, characterized in that: include: A data acquisition module, used to acquire user feature data and user operation data, wherein the user operation data includes search data and selection operations; a relationship determination module, configured to obtain label data based on the search data, and determine whether an item is selected based on the selection operation, wherein the item is an item searched by the search operation corresponding to the search data; A data processing module, configured to use the label data as positive sample label data of the item when the item is selected; and to extract a plurality of labels other than the label data from a label library as negative sample label data of the item when the item is selected; The model training module is used to obtain training data based on the positive and negative sample label data of the item, the item information, the user operation data and the user feature data, and based on the training data, use a machine learning algorithm to train the label matching detection model.

7. A label matching detection device, characterized in that: include: A model building module, used to build a label matching detection model according to the method described in any one of claims 1 to 3; The matching degree detection module is used to obtain the data to be tested and input it into the label matching degree detection model to obtain the matching degree between the object and the label in the data to be tested, wherein the data to be tested includes the label, user feature data, item information of the item and user operation data.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 5 is implemented.

9. An electronic device, characterized in that: include: processor; A memory, configured to store executable instructions of the processor; The processor is configured to perform the method of any one of claims 1 to 5 by executing the executable instructions.

Images