Recommended object style discrimination method and device, equipment and storage medium

Abstract

The invention discloses a recommended article style discrimination method and device, equipment and a storage medium, and belongs to the technical field of computers and Internet. Comprising the steps that attribute information of a first recommended object is acquired, and the attribute information of the first recommended object comprises a plurality of object attributes of the first recommended object; according to a category attribute rule and the attribute information of the first recommended object, identification information of the first recommended object is determined, the category attribute rule is a rule for determining the same type of recommended object under a first category to which the first recommended object belongs, and the identification information of the first recommended object is used for representing the first recommended object under the first category; according to the identification information of the first recommended object and the identification information of the second recommended object, a style judgment result is determined, the determination mode of the identification information of the second recommended object is the same as the determination mode of the identification information of the first recommended object, and the style judgment result is used for representing the style similarity between the first recommended object and the second recommended object. According to the scheme, the adaptability of the method for discriminating the style of the recommended object to different scenes can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computers and the Internet, and in particular to a method and device for determining the style of a recommended item, equipment and a storage medium. BACKGROUND

[0002] Item attributes are used to describe the attribute information of recommended items (such as brand, style, device parameters, etc.), and through item attributes, the same recommended items can be quickly found to complete price comparison, item recommendation, etc.

[0003] In related technologies, artificial or artificial intelligence models are used to determine the same recommended items on different platforms. For the method of using an artificial intelligence model to determine the same recommended items, the recommended item information of two recommended items is first obtained, the recommended item information of the two recommended items is spliced to obtain model input; then, the similarity between the recommended item information possessed by each of the two recommended items is determined through mechanisms such as the attention mechanism in the artificial intelligence model to obtain an output result; and it is determined whether the recommended item styles of the two recommended items are the same according to the output result.

[0004] However, interference factors exist in the recommended item information, etc., which causes this method to only guarantee a high accuracy of the style determination result in a specific scenario, and the generalization performance is poor. SUMMARY

[0005] The present application provides a method and device for determining the style of a recommended item, equipment and a storage medium. The technical solutions provided by the present application are as follows:

[0006] According to an aspect of an embodiment of the present application, a method for determining the style of a recommended item is provided, which comprises:

[0007] Obtaining attribute information of a first recommended item, wherein the attribute information of the first recommended item includes a plurality of item attributes possessed by the first recommended item;

[0008] Determining identification information of the first recommended item according to a category attribute rule and the attribute information of the first recommended item, wherein the category attribute rule is a same recommended item determination rule under a first category to which the first recommended item belongs, and the identification information of the first recommended item is used to represent the first recommended item under the first category;

[0009] Determining a style determination result according to the identification information of the first recommended item and the identification information of a second recommended item, wherein the determination manner of the identification information of the second recommended item is the same as that of the identification information of the first recommended item, and the style determination result is used to represent the style similarity between the first recommended item and the second recommended item.

[0010] According to an aspect of some embodiments of the present application, a device for determining a recommended item style is provided, and the device comprises:

[0011] an information obtaining module configured to obtain attribute information of a first recommended item, the attribute information of the first recommended item comprising a plurality of item attributes possessed by the first recommended item;

[0012] a label determining module configured to determine label information of the first recommended item according to a category attribute rule and the attribute information of the first recommended item, the category attribute rule being a same-item recommended item determination rule under a first category to which the first recommended item belongs, and the label information of the first recommended item being used to represent the first recommended item under the first category;

[0013] a result determining module configured to determine a style determination result according to the label information of the first recommended item and label information of a second recommended item, the label information of the second recommended item being determined in the same manner as the label information of the first recommended item, and the style determination result being used to represent a style similarity between the first recommended item and the second recommended item.

[0014] According to an aspect of some embodiments of the present application, a computer device is provided, and the computer device comprises a processor and a memory, the memory storing a computer program, the computer program being loaded and executed by the processor to implement the method for determining a recommended item style.

[0015] According to an aspect of some embodiments of the present application, a computer readable storage medium is provided, and the storage medium stores a computer program, the computer program being loaded and executed by a processor to implement the method for determining a recommended item style.

[0016] According to an aspect of some embodiments of the present application, a computer program product is provided, and the computer program product comprises a computer program, the computer program being stored in a computer readable storage medium, and a processor reading and executing the computer program from the computer readable storage medium to implement the method for determining a recommended item style.

[0017] The technical solutions provided by the embodiments of the present application have at least the following beneficial effects:

[0018] After obtaining the attribute information of the recommended items, the identification information possessed by the two recommended items is determined based on the attribute category rule, and the style discrimination result is determined through the identification information possessed by the two recommended items. Compared with directly using the recommended item information or the attribute information from the recommendation platform to determine whether the styles of the two recommended items are the same in the related art, the scheme provided in the embodiment of the present application uses the unified attribute category rule to correct the attribute information of the recommended items before determining the identification result of the recommended items, and obtains the identification information that can represent the style of the recommended items. On the one hand, the identification information is more concise than the attribute information, which helps to improve the calculation efficiency of determining the style discrimination result through the identification information. On the other hand, the identification information of the recommended items is obtained through the category attribute rule, which can exclude the differences in item attribute naming and the like of the recommendation platform, improve the adaptability of the style discrimination method of the present recommended item to cross-platform comparison needs, and improve the generalization ability of the method. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 is a schematic diagram of a computer system provided by an example embodiment of the present application;

[0020] Figure 2 is a flowchart of a style discrimination method of recommended items provided by an example embodiment of the present application;

[0021] Figure 3 is a schematic diagram of a category attribute system provided by an example embodiment of the present application;

[0022] Figure 4 is a schematic diagram of an entity recognition model provided by an example embodiment of the present application;

[0023] Figure 5 is a schematic diagram of a style discrimination method of recommended items provided by an example embodiment of the present application;

[0024] Figure 6 is a schematic diagram of a style discrimination method of recommended items in a comparison scenario;

[0025] Figure 7 is a block diagram of a style discrimination apparatus of recommended items provided by an example embodiment of the present application;

[0026] Figure 8 is a structural block diagram of a computer device provided by an example embodiment of the present application. DETAILED DESCRIPTION

[0027] To make the purpose, technical scheme and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the drawings.

[0028] First, the terms involved in the present application are explained.

[0029] Artificial Intelligence (AI) is the theory, method, technology and application system that use digital computers or digital computer controlled machines to simulate, extend and expand human intelligence, perceive environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology of computer science, which attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines, so that machines have the functions of perception, reasoning and decision-making.

[0030] Artificial intelligence technology is a comprehensive discipline, involving a wide range of fields, both hardware and software technologies. Artificial intelligence basic technologies generally include sensors, special artificial intelligence chips, cloud computing, distributed storage, big data processing technology, pre-training model technology, operation / interaction system, mechatronics, etc. Among them, the pre-training model, also known as the large model, the basic model, can be widely used in downstream tasks of various directions of artificial intelligence after fine-tuning. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology and machine learning / deep learning, etc.

[0031] Nature Language Processing (NLP) is an important direction in the field of computer science and artificial intelligence. It studies various theories and methods that can realize effective communication between people and computers using natural language. The research in this field will involve natural language, i.e. the language used in daily life, so it has a close relationship with the study of linguistics. Natural language processing technology usually includes text processing, semantic understanding, machine translation, robot question and answer, knowledge graph, etc.

[0032] Pre-training model, an important technology in artificial intelligence model training, is developed from Large Language Model (LLM) in the field of natural language processing. After fine-tuning, the large language model can be widely used in downstream tasks.

[0033] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithmic complexity theory, etc. It is a specialized study of how computers simulate or implement human learning behavior to acquire new knowledge or skills, reorganize existing knowledge structure, and continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental approach to making computers intelligent. It is applied in various fields of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and teaching learning.

[0034] The same recommended item refers to multiple recommended items that are objectively comparable and have the same important attributes. A recommended item refers to an item recommended to a user on an electronic recommendation platform. The terminal device acquires the right to use the recommended item or owns the recommended item in response to the user's recommended item acquisition operation.

[0035] With the research and progress of artificial intelligence technology, artificial intelligence technology is applied in many fields, such as common smart home, smart wearable device, virtual assistant, smart speaker, smart marketing, unmanned vehicle, autonomous vehicle, unmanned aerial vehicle, digital twin, virtual human, robot, artificial intelligence generated content, conversational interaction, intelligent medical treatment, intelligent customer service, game AI, etc.

[0036] It is believed that with the development of technology, artificial intelligence technology will be applied in more fields. For example, using an artificial intelligence model to determine the style of a recommended item on a recommendation platform, assisting staff on the recommendation platform to find the same recommended item, realizing cross-platform comparison of recommended item prices, and helping consumers quickly find the same recommended item. It has a wide range of application scenarios in the field of intelligent customer service and intelligent marketing.

[0037] Figure 1 is a schematic diagram of a computer system provided by an exemplary embodiment of the present application. The scheme implementation environment can include a terminal device 10, a server 20, and a computer device 30.

[0038] The terminal device 10 can be an electronic device such as a personal computer, a tablet computer, a mobile phone, a wearable device, a smart home appliance, a vehicle-mounted terminal, a virtual reality device, an augmented reality device, etc. The terminal device 10 can be a user of the recommended item style determination method. The terminal device 10 runs a client of a target application program. The target application program has the functions of recommending items, comparing prices of recommended items, etc. For example, the target application program has the function of automatically finding the same recommended item. For another example, the target application program is used to compare the styles of recommended items on different recommendation platforms.

[0039] In addition, the target application can also be a shopping application, a social application, an interactive entertainment application, a browser application, a content sharing application, a virtual reality application, an augmented reality application, etc., and the embodiments of the present application do not limit this. In addition, the type of recommended item is different for different applications, and the embodiments of the present application do not limit this.

[0040] There is a communication connection between the terminal device 10 and the server 20, and the identification process of the recommended item style is mainly completed by the server 20 in the embodiments of the present application.

[0041] The server 20 is used to provide background services for the client of the target application in the terminal device 10. For example, the server 20 can be a stand-alone physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks, (Content Delivery Network, CDN), and big data and artificial intelligence platforms, etc. Basic cloud computing services, but not limited to this.

[0042] The server 20 at least has data receiving function, storage and operation function.

[0043] In some embodiments, the category attribute rules are stored in the server 20, and the identification method of the recommended item style is realized based on the category attribute rules. For the determination method of the category attribute rules, please refer to the following embodiments.

[0044] After starting the identification process of the recommended item style, the server 20 obtains the attribute information of the first recommended item. The attribute information of the first recommended item includes a plurality of item attributes possessed by the first recommended item. Optionally, the attribute information of the first recommended item is obtained from the recommendation platform to which the first recommended item belongs, or the attribute information of the first recommended item is received by the terminal device 10, or the attribute information of the first recommended item is extracted by the server 20. For the specific content of this process, please refer to the following embodiments.

[0045] The server 20 determines the identification information of the first recommended item according to the above-mentioned category attribute rules and the attribute information of the first recommended item. The identification information of the first recommended item can represent the style of the first recommended item, and the style similarity between the first recommended item and other recommended items is determined by the identification information of the first recommended item and the identification information of the other recommended items.

[0046] Since the attribute information of the first recommended item includes multiple item attributes, the multiple item attributes have different representation capabilities for the style of the recommended item, in order to reduce the calculation amount in determining the style discrimination result, the server selects at least one item attribute from the multiple attribute information according to the category attribute rule, and determines the identification information of the first recommended item according to the at least one item attribute. The way in which the server 20 selects at least one item attribute from the multiple item attributes is determined based on the category attribute rule. That is, the server 20 determines which item attributes are selected from the multiple item attributes as the label information of the first recommended item based on the category attribute rule.

[0047] Optionally, the category attribute rule includes attribute distribution information, the attribute distribution information is used to represent the distribution of the item attributes in each style recommended item under the first category, and the server 20 selects at least one item attribute from the multiple item attributes according to the attribute distribution information; and splices the recommended item title of the first recommended item and the at least one item attribute to obtain the identification information of the first recommended item. For specific content of this process, please refer to the following embodiments.

[0048] After determining the identification information of the first recommended item in this way, the server 20 determines the style discrimination result based on the identification information of the first recommended item and the identification information of the second recommended item through the style discrimination model. Exemplarily, the style discrimination model is a model based on attention mechanism, and the server 20 generates input information of the style discrimination model according to the identification information of the first recommended item and the identification information of the second recommended item, and determines the context relationship based on the input information by the style discrimination model, so as to generate the style discrimination result.

[0049] Optionally, the category attribute rule includes attribute indication information, the attribute indication information is used to indicate at least one key attribute for distinguishing the styles of the recommended items under the first category. The server 20 determines at least one key attribute from the multiple item attributes according to the attribute indication information; and determines the at least one key attribute as the identification information of the first recommended item. That is, in this case, the server 20 directly selects the key attributes from the multiple item attributes according to the attribute indication information, and takes these key attributes as the identification information of the first recommended item.

[0050] Subsequently, the server 20 judges whether the key attributes of the first recommended item and the second recommended item are consistent, if the attribute contents of the first recommended item and the second recommended item in the at least one key attribute are consistent, it is determined that the style discrimination result is used to represent that the first recommended item and the second recommended item are same style recommended items; if the attribute contents of the first recommended item and the second recommended item in any one key attribute are inconsistent, it is determined that the style discrimination result is used to represent that the first recommended item and the second recommended item are not same style recommended items.

[0051] The computer device 30 runs at least one artificial intelligence model required in the process of determining the styles of the recommended items. The computer device 30 can be the same device as the server 20, or can be another device different from the server 20.

[0052] Optionally, in the case where the computer device 30 and the server 20 are different devices, there is a communication connection between the server 20 and the computer device 30, so that the server 20 transmits the model input of the artificial intelligence model to the computer device, and the computer device 30 feeds back the output result of the artificial intelligence model to the server 20.

[0053] The determination of the styles of the recommended items is generally used to determine whether two recommended items are the same recommended item. The same recommended item refers to any two recommended items with the same item attributes, that is, if the appearance and model of two recommended items are completely consistent, the two recommended items belong to the same recommended item. When manually determining the styles of the recommended items, the staff compares the recommended item parameters of the two recommended items and determines whether the two recommended items are the same recommended item based on work experience. With the development of artificial intelligence technology, in order to improve the determination efficiency of the same recommended item, there is currently a technology for using an artificial intelligence model to complete the determination of the same recommended item.

[0054] In the related art, after obtaining the recommended item information possessed by two recommended items from a recommended platform, the recommended item information possessed by the two recommended items is directly spliced to obtain a model input, the model input is directly processed by an artificial intelligence model to obtain a style determination result. Moreover, the corpus used in the training process of the artificial intelligence model is also directly obtained from the recommended platform.

[0055] There are various types of recommended items in the recommended item platform, and in order to more conveniently manage the recommended items, an attribute structure such as a category tree is usually used to manage different types of recommended items. The category tree includes multiple category branches. Optionally, there is no overlap between different category branches, and for any recommended item, the recommended item belongs to one of the multiple category branches. Illustratively, the category tree includes multiple nodes, and there is a parent-child node with a hierarchical structure between the multiple nodes. For any parent node in the category tree and the child node corresponding to the parent node, the parent node belongs to the upper node of the child node, and the parent node and the child node belong to the same category branch. The category corresponding to the child node is contained in the category corresponding to the parent node. One parent node can have multiple child nodes, and one child node has only one parent node. If two recommended items are the same recommended item, the categories to which the two recommended items belong are the same.

[0056] The category tree includes a plurality of leaf nodes, and the leaf node refers to a node without a child node in the category tree. Each leaf node corresponds to a minimum category, and the item attributes in the categories corresponding to different nodes can be the same or different. The item attributes are used to describe the recommended item from a specific perspective.

[0057] The structures of the category trees used by different recommendation platforms can be different, and the number and types of item attributes included in the categories corresponding to the leaf nodes with similar meanings in different category trees can also be different. That is, the modes of managing the recommended items are different for different recommendation platforms, and these differences have a non-negligible impact on the identification process of the recommended item styles, resulting in a low accuracy of the recommended item style identification method using an artificial intelligence model in related technologies.

[0058] In combination with the above-mentioned terms and application scenarios, the method for identifying the recommended item style provided by the present application is introduced, and the method is taken as an example applied to a computer device for illustration. As shown in Figure 2 The method can include the following steps (210-230).

[0059] Step 210: Obtain attribute information of a first recommended item.

[0060] The attribute information of the first recommended item includes a plurality of item attributes possessed by the first recommended item.

[0061] In some embodiments, the first recommended item is an item that can be purchased. Optionally, the first recommended item is a recommended item recommended by a recommendation platform. Exemplarily, the types of the first recommended item include but are not limited to goods in an e-commerce platform, recommended items in an advertisement, etc.

[0062] Optionally, the first recommended item includes but is not limited to at least one of the following: a physical recommended item (such as an electronic product, a daily consumable, a food, a pet, etc.), a service (such as a picture editing service, a video editing service, a device repair service, a membership service), a course (such as a course of a tutoring agency), etc.

[0063] In some embodiments, the first recommended item is any recommended item in the recommendation platform. Optionally, the first recommended item is specified by a terminal device. Optionally, in the scenario of identifying the same recommended item, the terminal device sends recommended item indication information to the computer device, and the recommended item indication information is used to indicate the first recommended item to the computer device. Exemplarily, the recommended item indication information includes a recommended item identifier of the first recommended item. The recommended item identifier can be one or more of a recommended item name, a recommended item tag, a recommended item picture, and a recommended item link. The recommendation platform includes an e-commerce platform, a sales website of the first recommended item, a social application platform for recommending the first recommended item, etc., and the present application does not limit the types of the recommended item and the recommendation platform.

[0064] In some embodiments, the item attribute of the first recommendation item is used to represent the type of the first recommendation item from a certain perspective. Optionally, the attribute information of the first recommendation item includes a plurality of item attributes, each of which is used to represent the first recommendation item from a different perspective. There is no item attribute of the same type in the plurality of item attributes included in the attribute information.

[0065] Optionally, the type of the item attribute includes, but is not limited to, at least one of the following: brand, model, content information, appearance information, parameter information, etc. The brand is used to indicate the manufacturer of the first recommendation item, the model is used to represent the rule of the first recommendation item, the content information is used to represent the quality, volume, capacity of effective content, etc. of the first recommendation item, the appearance information is used to represent the appearance of the first recommendation item, and the appearance information includes, but is not limited to, shape specification, color, material, etc. The parameter information is used to represent the structural composition of the first recommendation item.

[0066] For example, in the case of the first recommendation item being an electronic product, the parameter information can be subdivided into: battery capacity, network performance, screen size, camera pixel, storage capacity, etc.

[0067] In some embodiments, the item attribute includes the following two parts: attribute type and attribute content. For any item attribute, the attribute type of the item attribute is used to uniquely identify the item attribute. Exemplarily, the attribute type is represented by the name, number, etc. of the item attribute. The attribute content is used to represent the actual attribute value of the recommendation item under the attribute type, and the attribute content is also called attribute value. For example, in the case of the first recommendation item being an electronic product, the first recommendation item has a battery capacity of 2000 mA, the attribute type is battery capacity, and the attribute content is 2000 mA.

[0068] In some embodiments, the attribute information of the first recommendation item is obtained from the recommendation platform to which the first recommendation item belongs. The computer device crawls the attribute information of the first recommendation item from the recommendation platform to which the first recommendation item belongs. In some other embodiments, in the scenario of same-model recommendation item identification, the attribute information of the first recommendation item is sent by the terminal device to the computer device.

[0069] In view of the fact that the category trees used in different recommendation platforms are different, the types and quantities of the item attributes divided under the same category are different, the computer device obtains the recommendation information of the first recommendation item from the recommendation platform, and extracts the attribute information of the first recommendation item from the recommendation information based on the category attribute system. The category attribute system is used to indicate the types of the item attributes under the first category, wherein the first category refers to the category to which the first recommendation item belongs. Optionally, different styles of recommendation items are included in the first category. The computer device determines which item attributes are included in the first category through the category attribute system, determines the attribute content of the first recommendation item included in each of the multiple item attributes, and obtains the attribute information of the first recommendation item. For specific content, please refer to the embodiments below.

[0070] In some embodiments, the category attribute system is stored and managed in the form of a category tree. The category tree includes at least one node with a hierarchical structure, and for each leaf node in the category tree, the leaf node corresponds to an indivisible minimum category. Optionally, the leaf node corresponds to n item attributes, and n is a positive integer. The n item attributes are the item attributes that the recommendation items belonging to the category should have. That is, the category tree and the item attributes corresponding to each leaf node in the category tree form the category attribute system.

[0071] For example, in the method for determining the style of the recommendation item provided in the present application, the category attribute system is common to multiple recommendation platforms. The category attribute system is determined based on the classification and attribute distribution of the recommendation items in the multiple recommendation platforms. For details, please refer to the embodiments below. The categories and attributes in the category attribute system are summarized by the staff according to the category setting rules in the multiple recommendation platforms.

[0072] Figure 3 FIG. 3 is a schematic diagram of a category attribute system provided in an example embodiment of the present application. As shown in FIG. 3, the categories included in the category attribute system are managed in the form of a category tree 310, and the multiple item attributes included in the “mobile phone” category are shown as 320. Figure 3

[0073] In this way, in the process of obtaining the attribute information of the first recommendation item, the attribute information of the recommendation items in different recommendation platforms is unified, which helps to improve the credibility of the cross-platform determination of the style of the recommendation item.

[0074] In some embodiments, the first recommendation item belongs to a first category in the category tree, the category tree includes multiple categories, the category tree is used to classify and manage the massive data in the recommendation platform, and any two recommendation items with the same style belong to the same category. The category tree is set by the staff of the recommendation platform.

[0075] In step 220, the identification information of the first recommendation item is determined according to the category attribute rules and the attribute information of the first recommendation item.​

[0076] The category attribute rule is a same-model recommendation determination rule of a first category to which the first recommendation belongs.

[0077] In some embodiments, the category attribute rule is used to indicate the determination standard of whether two recommendations belong to the same model.

[0078] Optionally, the identification information of the first recommendation includes at least one of the plurality of attribute information. The computer device selects at least one attribute information from the plurality of attribute information according to the category attribute rule to obtain the identification information of the first recommendation.

[0079] In one example, the computer device directly selects at least one item attribute from the plurality of item attributes based on the category attribute rule. In another example, the computer device determines the selection priority of the plurality of item attributes based on the category attribute rule, and selects at least one item attribute from the plurality of item attributes based on the selection priority. For specific contents, please refer to the following embodiments.

[0080] In some embodiments, the first category is the category to which the first recommendation belongs. Optionally, the first category is a third-level category in the category tree, that is, the category corresponding to the leaf node in the category tree. Of course, the first category can also be the category corresponding to the parent node in the category tree, that is, the first category includes more types of recommendations. For example, in the case where a parent node corresponds to a plurality of child nodes and the category attribute rules of the plurality of child nodes are the same, the first category is the category corresponding to the parent node.

[0081] The identification information of the first recommendation is used to represent the first recommendation in the first category.

[0082] In some embodiments, the identification information of the first recommendation is used to uniquely represent the recommendation model of the first recommendation in the first category. Optionally, for two recommendations belonging to different recommendation models in the first category, the two recommendations each have different identification information. For example, in the case where the representation information includes item attributes, the difference in the identification information includes the following two cases: 1. The attribute types of the item attributes are different; 2. The attribute types of the item attributes are the same, and the attribute contents are different.

[0083] Optionally, the category attribute rule is determined in advance or in the process of distinguishing the recommendation model. Pre-determining the category attribute rule helps to improve the speed of distinguishing the recommendation model.

[0084] In some embodiments, the category attribute rule is determined based on a statistical frequency of the item attribute under the first category. Optionally, for the first category in the category attribute system, the computer device obtains a plurality of recommended items under the first category, the plurality of recommended items being from at least two recommendation platforms; for each of the plurality of recommended items, the computer device determines attribute information of the recommended item according to recommended item information possessed by the recommended item in a recommendation platform to which the recommended item belongs; the computer device classifies the plurality of recommended items to obtain at least one recommended item group, wherein for any two recommended items in the at least one recommended item group, the two recommended items have at least one same recommended item characteristic; the computer device excludes at least one recommended item from the recommended item group to obtain an excluded recommended item group, the excluded recommended item group including a plurality of recommended items belonging to the same style; subsequently, the computer device, taking the excluded recommended item group as a unit, counts a frequency of occurrence of the item attribute possessed by each recommended item to determine the category attribute rule. For specific calculation methods of the category attribute rule, please refer to the following embodiments.

[0085] Illustratively, the computer device obtains the plurality of recommended items under the first category, including: for any recommendation platform, the computer device performs category search on the platform to obtain the plurality of recommended items under the first category. For example, a mobile phone is a category, and the computer device searches for “mobile phone” in a plurality of recommendation platforms respectively, and takes the first 1000 search results as the plurality of recommended items.

[0086] Illustratively, the computer device determines the attribute information of the recommended item according to recommended item information possessed by the recommended item in a recommendation platform to which the recommended item belongs, including: processing the recommended item information of the recommended item by an entity recognition model to obtain the item attribute of the recommended item, the entity recognition model being based on training data sorted out under the category attribute system.

[0087] Illustratively, the computer device classifies the plurality of recommended items to obtain at least one recommended item group, including: classifying the plurality of recommended items according to a preposition rule to obtain at least one recommended item group. The preposition rule includes but is not limited to at least one of the following: consistent product batch number, consistent manufacturer name, consistent place of origin, etc. For example, for two cosmetic products, the computer device determines whether the product batch numbers of the two cosmetic products are consistent. If the product batch numbers of the two products are consistent, it is determined that the two cosmetic products belong to the same recommended item group; if the product batch numbers of the two products are not consistent, it is determined that the two cosmetic products belong to different recommended item groups.

[0088] Exemplarily, the computer device classifies the plurality of recommendations to obtain at least one recommendation group, including: for any two recommendations, determining embedding representations of the two recommendations based on recommendation information of the two recommendations, calculating similarity between the two embedding representations; if the similarity is greater than a similarity threshold, determining that the two recommendations belong to one recommendation group; if the similarity is less than or equal to the similarity threshold, determining that the two recommendations belong to different recommendation groups. The embedding representation is used to represent the content of the recommendation information in the form of vector information, that is, only the recommendation information is mapped to a feature space to obtain the corresponding embedding representation. The similarity between the embedding representations is represented by the vector distance between the two embedding representations. The similarity threshold is a pre-set value.

[0089] Exemplarily, the computer device screens out at least one recommendation from the recommendation group to obtain a screened recommendation group, including: in response to a label identification operation, setting a screening label for the recommendation to be screened out, and based on the screening label, screening out at least one recommendation to be screened out from the recommendation group. The label identification operation includes but is not limited to: click operation, key operation, sliding operation, long press operation, etc. on the recommendation to be screened out.

[0090] Step 230, determining a style discrimination result according to the identification information of the first recommendation and the identification information of the second recommendation.

[0091] The determination manner of the identification information of the second recommendation is the same as the determination manner of the identification information of the first recommendation, and the style discrimination result is used to represent the style similarity between the first recommendation and the second recommendation.

[0092] The second recommendation refers to a recommendation in a recommendation platform. Optionally, the second recommendation and the second recommendation belong to the same category. Exemplarily, the recommendation platform to which the first recommendation belongs and the recommendation platform to which the second recommendation belongs are different. Since the category attribute rule in the scheme provided in the present application is obtained based on the item attribute statistics of the recommendations in different recommendation platforms, the category attribute rule is applicable to the goods in different recommendation platforms.

[0093] In some embodiments, the determination manner of the identification information of the second recommendation is the same as the determination manner of the identification information of the first recommendation, that is, the computer device determines the identification information of the second recommendation according to the category attribute rule and the attribute information of the second recommendation. Optionally, the determination timing of the identification information of the second recommendation is prior to the determination timing of the first identification information, such as the identification information of the second recommendation is determined in advance. Of course, the determination timing of the identification information of the second recommendation can also be later than the determination timing of the first identification information, or the identification information of the first recommendation and the identification information of the second recommendation are determined synchronously, and the present application does not limit the determination timing.

[0094] In some embodiments, the style discrimination result is used to represent whether the first recommendation and the second recommendation are same-style recommendations. Optionally, the style discrimination result is a decimal number in the interval of [0, 1]. If the style discrimination result is greater than a matching threshold, it indicates that the first recommendation and the second recommendation are same-style recommendations; if the style discrimination result is less than or equal to the matching threshold, it indicates that the first recommendation and the second recommendation are not same-style recommendations.

[0095] Optionally, the style discrimination result is used to directly indicate whether the first recommendation and the second recommendation are same-style recommendations. Illustratively, the style discrimination result has two optional values. For example, the style discrimination result equal to 1 indicates that the first recommendation and the second recommendation are same-style recommendations, and the style discrimination result equal to 0 indicates that the first recommendation and the second recommendation are not same-style recommendations.

[0096] Optionally, the computer device determines the style discrimination result according to the matching degree of the identification information of the first recommendation and the identification information of the second recommendation. Illustratively, the style discrimination result is determined by comparing the overlapping degree between the identification information of the first recommendation and the identification information of the second recommendation. For example, the identification information of the recommendation includes at least one item attribute, and the style discrimination result is obtained by determining whether there is a consistent item attribute in the identification information of the first recommendation and the identification information of the second recommendation.

[0097] Optionally, the computer device determines the style discrimination result through a style discrimination model. The style discrimination model is used to determine the style discrimination result based on the identification information possessed by each of the two recommendations. Illustratively, the style discrimination model is an artificial intelligence model. The types of the style discrimination model include, but are not limited to, a Transformers model, a Bidirectional Encoder Representation from Transformers (BERT) model based on Transformers, a Simple Contrastive Learning of Sentence Embeddings (SimCSE) model for sentence embedding learning, etc. It should be noted that the type of the style discrimination model is selected according to actual needs, and is not limited herein.

[0098] Illustratively, a model input is generated based on the identification information of the first recommendation and the identification information of the second recommendation, the model input is transmitted to the style discrimination model, and the style discrimination result generated by the style discrimination model is obtained.

[0099] In some embodiments, in the case that the style discrimination model is used in the generation of the style discrimination result, the style discrimination model is obtained by fine tuning an initial style discrimination model based on the first training data under the same category attribute system.

[0100] In summary, after obtaining the attribute information of the recommended items, the identification information possessed by the two recommended items is determined based on the attribute category rules, and the style discrimination result is determined through the identification information possessed by the two recommended items. Compared with directly using the recommended item information or attribute information from the recommendation platform to determine whether the styles of the two recommended items are the same, the scheme provided in the embodiments of the present application uses the unified attribute category rules to correct the attribute information of the recommended items before determining the identification result of the recommended items, so as to obtain the identification information that can represent the style of the recommended items.

[0101] On the one hand, the identification information is more refined than the attribute information, which helps to improve the calculation efficiency of determining the style discrimination result through the identification information. On the other hand, the identification information of the recommended items is obtained through the category attribute rules, which can eliminate the differences in item attribute naming and the like in the recommendation platform, improve the adaptability of the style discrimination method of the recommended items to cross-platform comparison needs, and improve the generalization ability of the method.

[0102] The following describes the acquisition process of the attribute information of the recommended items through several embodiments.

[0103] It is considered that different recommendation platforms use different category trees, and for two recommended items of the same style but belonging to different recommendation platforms, the initial attribute information possessed by the two recommended items is also different, and the names of the recommended items of the same style set by different merchants in the recommendation platform are also not all the same, which will affect the style discrimination result. In order to improve the accuracy of the style discrimination result determined by the scheme of the present application, it is necessary to convert the recommended items to a common category attribute system, so as to realize the unification of the attributes of the recommended items, and facilitate the subsequent steps of the style discrimination process of the recommended items.

[0104] In some embodiments, as shown in step 210, the attribute information of the first recommended item is obtained, including the following sub-steps. Figure 2

[0105] Sub-step 211: obtaining the recommended item information of the first recommended item from the recommendation platform, the recommendation platform being used to provide recommended items to users.

[0106] In some embodiments, the recommended item information of the first recommended item refers to the description information of the first recommended item in the recommendation platform. Alternatively, the recommended item information of the first recommended item is the recommended item title displayed in the purchase page of the first recommended item in the recommendation platform. ​

[0107] In order to improve the recall efficiency of the recommended object, keywords related to the recommended object are often stacked in the title of the recommended object as much as possible to improve the coincidence of the title of the recommended object with the keywords of the user search, and improve the recall rate of the recommended object. In the present application, the recommended object information of the first recommended object is obtained from the recommendation platform by describing the characteristics of the recommended object from multiple angles by the title of the recommended object, so as to facilitate subsequent complete determination of the item attributes included in the first recommended object according to the recommended object information of the first recommended object.

[0108] For specific introduction of the recommendation platform, please refer to the above embodiments, which will not be repeated here. Exemplarily, in the same product recommended object index scene, the terminal device sends the identification information of the first recommended object to the computer device, the computer device finds the first recommended object from the recommendation platform according to the identification information of the first recommended object, and crawls the recommended object information of the first recommended object from the item page of the recommended object of the first recommended object.

[0109] In step 213, the recommended object information of the first recommended object is identified by the entity recognition model to obtain the attribute information of the first recommended object, and the entity recognition model is adjusted based on the second training data. The second training data is extracted based on the category attribute system from the recommended object information in at least one recommendation platform.

[0110] In some embodiments, the entity recognition model is used to complete the named entity recognition (NER) task. In the embodiments of the present application, the entity recognition model is used to extract a plurality of item attributes from the recommended object information. Optionally, the entity recognition model is a bidirectional long short term memory conditional random field (BI LSTM CRF) model.

[0111] In order to enable the entity recognition model to better extract item attributes from the recommended object information and obtain the attribute information of the recommended object, the model parameters of the initial entity recognition model need to be adjusted before using the entity recognition model to obtain the entity recognition model used in the present application. The step of adjusting the initial entity recognition model can be performed by the computer device, or by other devices other than the computer device.

[0112] Optionally, the second training data used in adjusting the initial entity recognition model is determined based on a category attribute system. That is, the second training data includes at least one item attribute from the category attribute system. During the adjustment of the initial entity recognition model, a mask is added to the item attributes to complete a cloze test using the initial recognition model, generating predicted attributes. A training loss is calculated based on the predicted attribute information and the item attributes. Model parameters are adjusted based on the training loss until the training loss converges to produce an entity recognition model.

[0113] Figure 4 It is a schematic diagram of an entity recognition model provided by an exemplary embodiment of the present application.

[0114] like Figure 4 As shown, the entity recognition model includes an embedding layer, an encoder, and a decoder. Optionally, identifying the recommended item information of the first recommended item through the entity recognition model to obtain the attribute information of the first recommended item includes: representing the attribute information of the first recommended item through the embedding layer to obtain an attribute embedding representation, which is used to represent the attribute information of the first recommended item in the form of a vector; transforming the attribute embedding representation through the encoder to obtain a feature code, which is used to represent the item attributes in the feature space; and decoding the feature code through the decoder to obtain the attribute information of the first recommended item.

[0115] Exemplarily, the attribute information of the first recommendation is represented by an embedding layer to obtain an attribute embedding representation, including: word segmentation processing of the attribute information of the first recommendation to obtain multiple word units, where a word unit refers to the smallest unit with independent semantics (such as a word in Chinese, a word in English); determining the word embedding representation of each word unit, and for the first word unit among the multiple word units, determining the auxiliary information of the first word unit from the attribute map; generating an embedding representation of the auxiliary information, splicing the word embedding identifiers and the embedding representations of the auxiliary information respectively possessed by multiple units, to obtain an attribute embedding representation.

[0116] The attribute graph refers to a knowledge graph related to item attributes, including auxiliary information for explaining the attributes. The auxiliary information of the first word unit is used to provide an explanation of the attributes of the first word unit, and the auxiliary information is text including at least one word unit.

[0117] This solution, based on an attribute category system, processes recommended item information from recommendation platforms to obtain attribute information. This standardizes the attributes of recommended items across different recommendation platforms and reduces errors in the style identification process caused by different attribute classification and naming methods. This helps improve the accuracy of style identification results and enhances the adaptability of this recommended item style identification method to different platforms and cross-platform style identification processes.

[0118] The determination process of the identification information of the recommended item is introduced below through several embodiments. The determination methods of the identification information of the recommended item in the embodiments of the present application mainly include three types, which are introduced below through three examples respectively.

[0119] Example 1

[0120] In some embodiments, the category attribute rule includes attribute distribution information, and the attribute distribution information is used to represent the distribution of the item attribute in each style recommended item under the first category.

[0121] Optionally, the category attribute rule is used to represent the representation ability of the item attribute possessed by the first category to the style of the recommended item. The category attribute rule is obtained by counting the item attributes respectively possessed by the recommended items of multiple recommendation platforms. For the determination method of the category attribute rule, please refer to the embodiments below.

[0122] Exemplarily, for each of the multiple item attributes, the distribution of the item attribute in each style recommended item under the first category is related to the attribute content of the item attribute. As introduced above, the item attribute includes the attribute type and the attribute content, and the same item attribute refers to the same attribute type and the same attribute content or different attribute contents. If the attribute contents of the multiple recommended items of the same style on the same item attribute are all the same, it indicates that the item attribute can significantly represent the style of the recommended item; if the attribute contents of the multiple recommended items of the same style on the item attribute are not all the same, it indicates that even the attribute contents of the recommended items of the same style on the item attribute are not completely consistent, and the representation ability of the item attribute to the style is weak.

[0123] In some embodiments, the category attribute rule includes the distribution information respectively possessed by the multiple item attributes, and for any item attribute, the distribution information of the item attribute is used to represent the distribution of the item attribute in each style recommended item under the first category. Optionally, in the process of determining the identification information of the first recommended item according to the category attribute rule and the attribute information of the first recommended item, the computer device selects at least one item attribute from the multiple item attributes through the distribution information respectively possessed by the multiple item attributes, and obtains the identification information of the first recommended item. For details, please refer to the embodiments below.

[0124] Figure 3 The step 220 in the method 200, determining the identification information of the first recommended item according to the category attribute rule and the attribute information of the first recommended item, includes the following sub-steps.

[0125] Sub-step 221, selecting at least one item attribute from the multiple item attributes according to the attribute distribution information.

[0126] Optionally, for each of the at least one item attribute (i.e., the selected item attribute), the item attribute has a stronger representation capability of the recommended item style under the first category than the unselected item attribute in the plurality of item attributes.

[0127] The process of selecting the at least one item attribute is described below through several embodiments.

[0128] In some embodiments, the sub-step 221 of selecting the at least one item attribute from the plurality of item attributes according to the attribute distribution information comprises the following sub-steps.

[0129] The sub-step 221a comprises determining a selection priority of a first item attribute in the plurality of item attributes according to distribution information of the first item attribute in the attribute distribution information.

[0130] The selection priority is used to represent the priority of selecting the item attribute from the plurality of item attributes, and the distribution information of the first item attribute is used to represent the distribution of the first item attribute in the plurality of recommended items under the first category, wherein the recommended items are from at least one recommendation platform, and the recommendation platform is used to provide the recommended items to the user.

[0131] The first item attribute refers to any one of the plurality of item attributes. For example, the first item attribute is an item attribute whose selection priority is not determined.

[0132] In some embodiments, the selection priority of the item attribute is determined in the process of determining the recommended item style. In other embodiments, the selection priority of the item attribute is determined before the process of determining the recommended item style starts, because the category attribute rule is a statistical result of the item attributes of the recommended items in the plurality of recommendation platforms, i.e., the category attribute rule can be obtained before the process of determining the recommended item style starts.

[0133] Optionally, after the category attribute rule is determined, the computer device determines the selection priority of each item attribute in the category attribute rule based on the distribution information of each item attribute. That is, in order to improve the determination efficiency of the recommended item style. In the sub-step 221a, the step of determining the selection priority of the item attribute can be performed in advance. In this way, it helps to reduce the number of times of determining the selection priority of the same item attribute when the method of determining the recommended item style is repeatedly executed, and helps to reduce unnecessary calculations. For the determination step of the selection priority, please refer to the following embodiments.

[0134] Optionally, the selection priority of the first item attribute is proportional to the priority of selecting the item attribute from the item attributes. The higher the selection priority of the first item attribute, the greater the priority of selecting the first item attribute from the item attributes, that is, the first item attribute is preferentially selected from the plurality of item attributes. The lower the selection priority of the first item attribute, the lower the priority of selecting the first item attribute from the item attributes, that is, other item attributes are preferentially selected from the plurality of item attributes, rather than the first item attribute.

[0135] Sub-step 221b: selecting the top k item attributes with the highest selection priority from the plurality of item attributes to obtain at least one item attribute, k being a positive integer.

[0136] For example, k is equal to 5, and the top 5 item attributes with the highest selection priority are selected from the plurality of item attributes. Illustratively, k is pre-set. For example, k is less than the total number of attributes of the plurality of item attributes.

[0137] Based on the selection priority, at least one item attribute is selected, and item attributes with strong style representation capabilities are preferentially selected, which helps to improve the representation accuracy of the identification information of the recommended item on the style of the recommended item, helps to reduce the amount of calculation in the style identification process compared to using complete attribute information for style identification, and helps to reduce the impact of irrelevant attributes on the accuracy of the style identification result.

[0138] Sub-step 223: splicing the recommended item title of the first recommended item and the at least one item attribute to obtain the identification information of the first recommended item.

[0139] In some embodiments, the recommended item title of the first recommended item is used to identify the first recommended item. Optionally, the recommended item title of the first recommended item includes but is not limited to the recommended item name of the first recommended item, the model number of the first recommended item, etc. Illustratively, the recommended item title of the first recommended item is obtained by crawling from the recommended platform to which the first recommended item belongs.

[0140] Since a recommended item has many item attributes, if the style identification process is directly completed based on a large number of item attributes, a large amount of calculation is required. In this step, at least one item attribute is selected from the plurality of item attributes, and item attributes with weak style representation capabilities are removed, and the identification information of the recommended item is generated based on the at least one item attribute, which helps to reduce the data amount of the identification information of the recommended item and helps to improve the determination efficiency of the style identification result.

[0141] The distribution information of the first item attribute is described below through several embodiments.

[0142] In some embodiments, the distribution information of the first item attribute comprises at least one of: a total frequency, a first frequency, a second frequency, a third frequency, and a fourth frequency.

[0143] The total frequency is the frequency of the first item attribute in the recommended items of each style under the first category; the first frequency is the frequency of the attribute content of the first item attribute being the same in the same-style recommended items under the first category; the second frequency is the frequency of the attribute content of the first item attribute being different in the same-style recommended items under the first category; the third frequency is the frequency of the attribute content of the first item attribute being the same in the different-style recommended items under the first category; and the fourth frequency is the frequency of the attribute content of the first item attribute being different in the different-style recommended items under the first category.

[0144] In some embodiments, the total frequency is used to represent the frequency of the first item attribute in each recommended item under the first category. Optionally, the total frequency is equal to the ratio of the total number of recommended items having the first item attribute under the first category to the total number of recommended items included under the first category. For example, the first item attribute is the memory capacity, the total number of recommended items included under the first category is equal to 10,000, and the total frequency is equal to the total number of recommended items having the first item attribute under the first category is equal to 8,000, then the total frequency of the first item attribute is equal to 80%.

[0145] In some embodiments, the first frequency is used to represent the proportion of the attribute content of the first item attribute being consistent in the same-style recommended items. Optionally, the more the number of recommended items with the attribute content of the first item attribute being consistent in the same-style recommended items, the higher the first frequency; the less the number of recommended items with the attribute content of the first item attribute being consistent in the same-style recommended items, the lower the first frequency. Illustratively, the larger the value of the first frequency, the stronger the representation ability of the first item attribute to the style, and the smaller the value of the first frequency, the weaker the representation ability of the first item attribute to the style.

[0146] Optionally, the computer device calculates a first sub-frequency corresponding to each style under the first category respectively; and determines the first frequency according to the first sub-frequency corresponding to each style respectively.

[0147] Illustratively, the computer device sums up the first sub-frequency corresponding to each style respectively based on the total number of recommended items included by each style respectively to obtain an intermediate value, and obtains the first frequency by dividing the intermediate value by the total frequency.

[0148] For example, under the first category, there are two styles, style 1 and style 2. In style 1, the number of recommended items with the same attribute content of the first item attribute is equal to 100, and the total number of recommended items included in style 1 is equal to 125, then the first sub-frequency corresponding to style 1 is equal to 80%. In style 2, the number of recommended items with the same attribute content of the first item attribute is equal to 90, and the total number of recommended items included in style 1 is equal to 100, then the first sub-frequency corresponding to style 2 is equal to 90%. The intermediate value is equal to 80%*125 / 225+90%*100 / 225=84%.

[0149] In some embodiments, the second frequency is used to represent the proportion of the attribute content of the first item attribute that is inconsistent in the same style recommended item. Optionally, the sum of the first frequency and the second frequency is equal to 1, and after the first frequency is calculated, the second frequency is equal to 1 minus the first frequency.

[0150] In some embodiments, the third frequency is used to represent the proportion of the attribute content of the first item attribute that is consistent in non-same style recommended items. Optionally, the third frequency is used to represent the credibility of the first item attribute in the recommended item style identification process. The greater the third frequency, the more consistent the attribute content of the first item attribute appears in different styles of recommended items, indicating that the credibility of the first item attribute in the recommended item style identification process is lower. The smaller the third frequency, the more different the attribute content of the first item attribute in different styles, indicating that the credibility of the first item attribute in the recommended item style identification process is higher.

[0151] For example, the third frequency is equal to the ratio between the frequency of the same attribute content in different styles of recommended items and the total frequency.

[0152] In some embodiments, the fourth frequency is used to represent the proportion of the attribute content of the first item attribute that is inconsistent in non-same style recommended items. Optionally, the fourth frequency is inversely proportional to the third frequency, and the sum of the third frequency and the fourth frequency is equal to 1.

[0153] By determining multiple comments of the item attribute, it is helpful to evaluate the ability of the item attribute to represent the style from multiple aspects, so as to select the item attribute with strong style representation ability from multiple item attributes, and determine the style discrimination result, which helps to improve the credibility of the style discrimination result.

[0154] The following describes the determination process of the selection priority through several embodiments.

[0155] In some embodiments, sub-step 221b, according to the distribution information of the first item attribute in the attribute distribution information, the selection priority of the first item attribute is determined, including the following sub-steps.

[0156] Sub-step 221b-10, determining the first weight according to the total frequency.

[0157] Optionally, the first weight is a positive number. Illustratively, the value of the first weight is equal to the total frequency.

[0158] Sub-step 221b-20, determining the second weight according to the first frequency and the fourth frequency, the influence of the total frequency on the first weight being greater than the influence of the first frequency and the fourth frequency on the second weight.

[0159] Optionally, calculating the sum of the first frequency and the fourth frequency to obtain a first result, and reducing the first result by a% to obtain the second weight, a being a positive number.

[0160] Sub-step 221b-30, determining the third weight according to the second frequency and the third frequency, the influence of the first frequency and the fourth frequency on the second weight being greater than the influence of the second frequency and the third frequency on the third weight.

[0161] Optionally, calculating the sum of the second frequency and the third frequency to obtain a second result, and reducing the second result by b% to obtain the second weight, b being a positive number smaller than a.

[0162] Sub-step 221b-40, determining the selection priority of the first item attribute according to the first weight, the second weight, and the third weight.

[0163] Optionally, the computer device adds the first weight, the second weight, and the third weight to obtain the selection priority of the first item attribute.

[0164] By calculating the selection priority of the item attribute through multiple frequencies, the style representation ability of the item attribute is measured from multiple aspects, which helps to improve the accuracy of the determined selection priority, and helps to improve the accuracy of the style recognition result by selecting the item attribute with better style representation ability according to the selection priority.

[0165] In some other embodiments, sub-step 221b, determining the selection priority of the first item attribute according to the distribution information of the first item attribute in the attribute distribution information, can be implemented as: ranking the multiple item attributes according to the total frequency to obtain an item attribute sequence; if the total frequencies of at least two item attributes in the multiple item attributes are equal, calculating the second weight of the at least two item attributes respectively, adjusting the positions of the at least two item attributes in the item attribute sequence based on the second weight; for any item attribute, determining the selection priority of the item attribute based on the position of the item attribute in the item attribute sequence.

[0166] Optionally, the second weight is related to the first frequency and the fourth frequency. For the calculation method of the second weight, please refer to the above embodiments, which will not be described here.

[0167] Exemplarily, the computer device sorts the plurality of item attributes in the order from high to low of the intermediate frequencies, to obtain an item attribute sequence; there are a first item attribute and a second item attribute with equal total frequencies in the plurality of item attributes, and then a second weight of the first item attribute and a second weight of the second item attribute are respectively calculated; assuming that the second weight of the first item attribute is greater than or equal to the second weight of the second item attribute, then in the item attribute sequence, the first item attribute is adjusted to be before the second item attribute, that is, the selection priority of the first item attribute is determined to be adjacent to the selection priority of the second item attribute, and the selection priority of the first item attribute is higher than the selection priority of the second item attribute.

[0168] Example 2:

[0169] In some embodiments, the category attribute rule includes attribute indication information, the attribute indication information is used to indicate at least one key attribute for distinguishing the recommended item style under the first category, and the key attribute is used to represent the recommended item style of the recommended item.

[0170] Optionally, the key attribute can directly represent the style of the recommended item. Exemplarily, the at least one key attribute is pre-set according to the experience of the staff.

[0171] Figure 2 As shown in step 220, according to the category attribute rule and the attribute information of the first recommended item, the identification information of the first recommended item is determined, including the following sub-steps.

[0172] In sub-step 225, according to the attribute indication information, at least one key attribute is determined from the plurality of item attributes.

[0173] Optionally, the attribute indication information includes the attribute identifier of the at least one key attribute. The attribute identifier is used to represent the attribute type in the key attribute. The attribute identifier includes but is not limited to the attribute name, label, etc. of the key attribute. Exemplarily, the computer device determines that the at least one key attribute has the attribute identifier according to the attribute indication information; the computer device determines at least one item attribute with the attribute identifier from the plurality of item attributes, and determines the at least one recommended item attribute as the at least one key attribute.

[0174] In sub-step 227, the at least one key attribute is determined as the identification information of the first recommended item.

[0175] For example, the attribute information of the first recommended item includes attribute 1, attribute 2, attribute 3 and attribute 4, and the key attribute includes attribute 1 and attribute 3, and then the identification information of the first recommended item is attribute 1 and attribute 3.

[0176] In this way, the identification information of the recommended product is quickly determined, the calculation pressure in the style determination process of the recommended product is reduced, and the efficiency of determining the style determination result is improved.

[0177] Example 3:

[0178] In some embodiments, the category attribute rule includes attribute distribution information and attribute indication information. For details of the attribute distribution information and the attribute indication information, please refer to the above embodiments.

[0179] In some embodiments, Figure 2 In step 220, the identification information of the first recommended product is determined according to the category attribute rule and the attribute information of the first recommended product, which includes the following sub-steps.

[0180] According to the attribute distribution information, at least one item attribute is selected from the plurality of item attributes. For example, at least one item attribute is selected from the plurality of item attributes according to the selection priority of the item attribute. The first recommended product title and the at least one item attribute are spliced to determine the first sub-identification information of the first recommended product. According to the attribute indication information, at least one key attribute is determined from the plurality of item attributes. The at least one key attribute is determined as the second sub-identification information of the first recommended product. The identification information of the first recommended product includes the first sub-identification information of the first recommended product and the second sub-identification information of the first recommended product. For details of this example, please refer to the above embodiments, which will not be repeated here.

[0181] The determination process of the style determination result will be introduced through several embodiments.

[0182] In some embodiments, step 230, the style determination result is determined according to the identification information of the first recommended product and the identification information of the second recommended product, which includes the following sub-steps.

[0183] Sub-step 231, the input information of the style determination model is determined according to the identification information of the first recommended product and the identification information of the second recommended product. The style determination model is trained based on the first training data of the category attribute system, and the category attribute system is used to classify the recommended products in the recommended platform.

[0184] Optionally, the input information of the style determination model includes the recommended product title of the first recommended product, the identification information of the first recommended product, the recommended product title of the second recommended product, and the identification information of the second recommended product. For example, in the case that the identification information of the first recommended product is at least one item attribute in the attribute information of the first recommended product, the input information of the style determination model includes at least one attribute of the first recommended product and at least one attribute of the second recommended product.

[0185] Exemplarily, in a case that the identification information of the first recommended product includes p product attributes, p is a positive integer greater than 1, the p product attributes form a first attribute sequence, and a product attribute with the highest priority is selected in front of the first attribute sequence. Exemplarily, in a case that the identification information of the second recommended product includes q product attributes, q is a positive integer greater than 1, the q product attributes form a second attribute sequence, and a product attribute with the highest priority is selected in front of the second attribute sequence.

[0186] Exemplarily, in a case that the identification information of the first recommended product includes a recommended product title of the first recommended product, the recommended product title of the first recommended product is located in front of any product attribute of the first recommended product in the input information of the style discrimination model. Exemplarily, in a case that the identification information of the second recommended product includes a recommended product title of the second recommended product, the recommended product title of the second recommended product is located in front of any product attribute of the second recommended product in the input information of the style discrimination model.

[0187] For example, the input information of the style discrimination model is represented as: [CLS] recommended product title of the first recommended product [SEP] identification information of the first recommended product [SEP] attribute information of the second recommended product [SEP] identification information of the second recommended product [SEP].

[0188] In some embodiments, the style discrimination model is used to determine a style discrimination result based on the identification information of the first recommended product and the identification information of the second recommended product. For specific introduction of the style discrimination model and the category attribute system, please refer to the above embodiments.

[0189] In some embodiments, the first training data includes: identification information of a first sample recommended product, identification information of a second sample recommended product, and a sample label. The identification information of the first sample recommended product and the identification information of the second sample recommended product each include at least one product attribute and a recommended product title. Optionally, the product attribute is a product attribute included in the category attribute system.

[0190] Optionally, the sample label is used to represent an actual style result between the first sample recommended product and the second sample recommended product. That is, the sample label is used to represent whether the first sample recommended product and the second sample recommended product actually belong to the same style recommended product. Exemplarily, the sample label is manually labeled, or the similarity between the recommended product title of the first sample recommended product and the recommended product title of the second sample recommended product is calculated by a text similarity model.

[0191] The adjustment process of the style discrimination model can be completed on a computer device or other devices, which is not limited herein. Optionally, the computer device generates input information of the initial style discrimination model based on the identification information of the first sample recommendation object and the identification information of the second sample recommendation object, processes the input information through the initial style discrimination model to obtain a predicted style result, calculates a training loss according to the predicted style result and the sample label, adjusts the model parameters of the initial style discrimination model based on the training loss until the training loss converges, and obtains the trained style discrimination model.

[0192] In some embodiments, the identification information of the first recommendation object includes at least one of the plurality of item attributes, and the at least one of the item attributes has a selection priority. For details of the selection priority, please refer to the above description.

[0193] In sub-step 231, input information of the style discrimination model is determined according to the identification information of the first recommendation object and the identification information of the second recommendation object, including the following steps.

[0194] In sub-step 231a, at least one item attribute included in the identification information of the first recommendation object is sorted according to the selection priority to obtain first input sub-information.

[0195] In sub-step 231b, if the character length of the first input sub-information is greater than a length threshold, i item attributes are truncated from the end of the first input sub-information, and i is a positive integer.

[0196] Optionally, the length threshold is less than or equal to half of the maximum input character length of the style discrimination model. For example, if the maximum input character length of the style discrimination model is 512, the length threshold is 256.

[0197] For example, if the character length of the first input sub-information is greater than the length threshold, the computer device deletes at least one item attribute with the lowest selection priority from the identification information of the first recommendation object.

[0198] In sub-step 231c, at least one item attribute included in the identification information of the second recommendation object is sorted according to the selection priority to obtain second input sub-information.

[0199] In sub-step 231d, if the character length of the second input sub-information is greater than the length threshold, j item attributes are truncated from the end of the second input sub-information, and j is a positive integer.

[0200] In sub-step 231e, the first input sub-information and the second input sub-information are spliced to obtain the input information of the style discrimination model.

[0201] In sub-step 233, the input information is processed through the style discrimination model to obtain a style discrimination result.

[0202] Figure 5 is a schematic diagram of a recommended item style identification method provided by an example embodiment of the present application.

[0203] As shown in Figure 5 , the computer device obtains recommended item information 511 of a first recommended item and recommended item information 512 of a second recommended item, processes the recommended item information 511 of the first recommended item and the recommended item information 512 of the second recommended item respectively through an entity recognition model to obtain attribute information of the first recommended item and attribute information of the second recommended item, then selects at least one item attribute from the attribute information of the first recommended item to obtain identification information of the first recommended item, and selects at least one item attribute from the attribute information of the second recommended item to obtain identification information of the first recommended item. The identification information of the first recommended item and the identification information of the second recommended item are used to generate input information 513 of a style identification model, the input information is processed through the style identification model to obtain a style identification result between the first recommended item and the second recommended item.

[0204] In some embodiments, the identification information includes at least one key attribute used to represent the style of the recommended item; and the step 230 of determining the style identification result according to the identification information of the first recommended item and the identification information of the second recommended item includes: if the attribute contents of the first recommended item and the second recommended item are consistent in at least one key attribute, it is determined that the style identification result is used to represent that the first recommended item and the second recommended item are the same recommended item.

[0205] Optionally, if for each key attribute included in the identification information of the first recommended item, there is a corresponding key attribute in the identification information of the second recommended item, and the two key attributes are consistent in attribute type and attribute content, it is determined that the style identification result is used to represent that the first recommended item and the second recommended item are the same recommended item; if for a certain key attribute included in the identification information of the first recommended item, there is no corresponding key attribute in the identification information of the second recommended item, or the key attributes are inconsistent in attribute content, it is determined that the style identification result is used to represent that the first recommended item and the second recommended item are not the same recommended item.

[0206] In this way, the recommended item style identification process can be efficiently completed.

[0207] In some embodiments, after the step of determining the style identification result according to the identification information of the first recommended item and the identification information of the second recommended item, the method further includes: obtaining a recommended item picture of the first recommended item and a recommended item picture of the second recommended item; determining a picture similarity according to the recommended item picture of the first recommended item and the recommended item picture of the second recommended item; and updating the style identification result according to the picture similarity to obtain an updated style identification result.

[0208] Figure 6 is a schematic diagram of a commodity style discrimination method in a price comparison scenario. In the present scheme, the commodity style discrimination method is triggered by a user terminal, and the server provides background support for the commodity style discrimination process. The following steps are executed by the server.

[0209] Step 610, obtaining a commodity price comparison request sent by a terminal device.

[0210] The commodity price comparison request is used to request to determine the price information of the to-be-compared commodity in at least one commodity recommendation platform.

[0211] Optionally, the commodity price comparison request includes at least one of the following: a commodity identifier of the to-be-compared commodity, a device identifier of the user terminal. The to-be-compared commodity can be a commodity in an electronic commodity selling platform, and the to-be-compared commodity corresponds to the first recommended object in the above embodiment.

[0212] Illustratively, the device identifier of the user terminal is used to uniquely represent the user terminal. The device identifier includes but is not limited to the device serial number of the user account, the network account of the user terminal, etc. Illustratively, the commodity price comparison request further includes a platform identifier, which is used to indicate at least one commodity recommendation platform that the user expects to compare prices. Correspondingly, the terminal device determines the commodity identifier of the to-be-identified commodity in response to an operation of triggering a price comparison behavior for the to-be-identified commodity, and generates a commodity price comparison request.

[0213] Illustratively, the operation of the price comparison behavior for the to-be-compared commodity is responded by a target application program. The target application program can be a commodity recommendation platform to which the first commodity belongs, or a separate application program specially used for price comparison.

[0214] Step 620, obtaining commodity information of the to-be-compared commodity from a first commodity recommendation platform according to the commodity identifier of the to-be-compared commodity. Optionally, the first commodity recommendation platform is used to recommend the to-be-compared commodity.

[0215] Step 630, identifying the commodity information of the to-be-compared commodity through an entity recognition model to obtain attribute information of the to-be-compared commodity.

[0216] The attribute information of the to-be-compared commodity includes a plurality of commodity attributes of the to-be-compared commodity. The commodity attribute corresponds to the item attribute in the above embodiment. In some embodiments, the commodity attribute includes an attribute type and an attribute content under the attribute type. The plurality of commodity attributes are used to describe the to-be-compared commodity from different angles. Optionally, the entity recognition model is obtained by adjusting an initial entity recognition model using training data. The training data includes commodity attributes, which are commodity attributes included in a category attribute system commonly used in a plurality of commodity recommendation platforms. For specific content of the category attribute system, please refer to the above embodiment.

[0217] At step 640, at least one key attribute of the to-be-priced commodity is determined from the plurality of commodity attributes according to the attribute indication information, and the at least one key attribute of the to-be-priced commodity is determined as the identification information of the to-be-priced commodity.

[0218] Optionally, the category attribute rule comprises attribute indication information, the attribute indication information is used to indicate at least one key attribute for distinguishing the commodity style under the first category, and the key attribute is used to represent the commodity style of the commodity; and the first category refers to a category to which the to-be-priced commodity belongs.

[0219] At step 650, commodity information of a second commodity is obtained from the commodity recommendation platform for each of the at least one commodity recommendation platform.

[0220] The second commodity corresponds to the second recommended item in the above embodiment. Optionally, the second commodity has at least one same search term as the to-be-priced commodity. The search term comprises at least one commodity attribute possessed by the to-be-priced commodity. For example, the search term related to the to-be-priced commodity is searched in the commodity recommendation platform, at least one second commodity is obtained from the search result, and commodity information of the second commodity is obtained by entering a recommendation page of the second commodity.

[0221] At step 660, attribute information of the second commodity is obtained by identifying the commodity information of the second commodity through the entity recognition model.

[0222] At step 670, at least one key attribute of the second commodity is determined from the attribute information of the second commodity according to the attribute indication information, and the at least one key attribute of the second commodity is determined as the identification information of the second commodity.

[0223] At step 680, if the attribute content of the to-be-priced commodity and the second commodity in the at least one key attribute is consistent, it is determined that the style distinguishing result is used to represent that the to-be-priced commodity and the second commodity are the same commodity. Subsequently, the following steps are skipped, and step 6120 is directly executed.

[0224] Optionally, if the attribute content of the to-be-priced commodity and the second commodity in the at least one key attribute is inconsistent, the identification information of the to-be-priced commodity and the identification information of the second commodity are re-determined from step 690.

[0225] At step 690, p commodity attributes are selected from a plurality of commodity attributes possessed by the to-be-priced commodity according to the attribute distribution information, the commodity title of the to-be-priced commodity and the p commodity attributes are spliced, and the identification information of the to-be-priced commodity is obtained, p being a positive integer.

[0226] Optionally, the category attribute rules include attribute distribution information, which is used to represent the distribution of product attributes among the various product styles under the first category. For example, for any product attribute, the attribute distribution information includes the distribution information of the product attribute. For details about step 690, please refer to the above embodiment.

[0227] Step 6100: Select q product attributes from the multiple product attributes of the second product according to the attribute distribution information, and combine the product title of the price comparison product and the q product attributes to obtain identification information of the second product, where q is a positive integer.

[0228] Step 6110: Determine input information of a style discrimination model based on the identification information of the product to be compared and the identification information of the second product, and process the input information through the style discrimination model to obtain a style discrimination result.

[0229] Step 6120: If the style identification result indicates that the style of the product to be identified is the same as that of the first product, price information of the second product is obtained from the product recommendation platform.

[0230] Step 6130: If the acquired price information is less than the quantity threshold, the step of acquiring the product information of the second product from each of the at least one product recommendation platform is repeated.

[0231] Step 6140: If the product price information is greater than or equal to the quantity threshold, price information related to the style of the product to be compared in at least one product recommendation platform is fed back to the user terminal.

[0232] In the scenario of price comparison of commodities applied in the embodiment of the present application, since the attribute information of commodities is determined based on the attribute category system common to multiple commodity recommendation platforms, and the category attribute rules in the process of determining the identification information of commodities based on the attribute information of commodities are also obtained based on the distribution statistics of commodity attributes in multiple commodity recommendation platforms, the solution provided by the present application has good adaptability to the price comparison needs between different commodity recommendation platforms. In the field of e-commerce sales, commodity management (commodity deduplication, information fusion, etc.) and price reference are very necessary, which is related to the sales advantage and information control of one's own commodities. The solution provided by the present application improves the accuracy of style discrimination results, helps to timely understand the price change trends of the same commodity on different commodity recommendation platforms, and helps to improve the competitiveness of the platform.

[0233] The following are device embodiments of the present application, which can be used to implement the method embodiments of the present application. For details not disclosed in the device embodiments of the present application, please refer to the method embodiments of the present application.

[0234] Figure 7A block diagram of a device for identifying a recommended item style is shown. The device can be implemented by software, hardware or a combination of both to form all or part of the device for identifying a recommended item style. The device 700 can include an information obtaining module 710, an identification determining module 720 and a result determining module 730.

[0235] The information obtaining module 710 is configured to obtain attribute information of a first recommended item, the attribute information of the first recommended item including a plurality of item attributes possessed by the first recommended item.

[0236] The identification determining module 720 is configured to determine identification information of the first recommended item according to a category attribute rule and the attribute information of the first recommended item, the category attribute rule being a same-item recommended item determining rule under a first category to which the first recommended item belongs, and the identification information of the first recommended item being used to represent the first recommended item under the first category.

[0237] The result determining module 730 is configured to determine a style identification result according to the identification information of the first recommended item and identification information of a second recommended item, the second recommended item having the same determination manner as the first recommended item, and the style identification result being used to represent a style similarity between the first recommended item and the second recommended item.

[0238] In some embodiments, the category attribute rule includes attribute distribution information, the attribute distribution information being used to represent a distribution of the item attributes in each style recommended item under the first category; and the identification determining module 720 includes an attribute selecting unit configured to select at least one item attribute from the plurality of item attributes according to the attribute distribution information, and an identification determining unit configured to splice a recommended item title of the first recommended item and the at least one item attribute to obtain the identification information of the first recommended item.

[0239] In some embodiments, the attribute selecting unit is configured to determine a selection priority of a first item attribute in the plurality of item attributes according to distribution information of the first item attribute in the attribute distribution information, the selection priority being used to represent a priority of selecting the item attribute from the plurality of item attributes, the distribution information of the first item attribute being used to represent a distribution of the first item attribute in a plurality of recommended items under the first category, the plurality of recommended items being from at least one recommendation platform, and the recommendation platform being used to provide recommended items to a user; and the at least one item attribute is obtained by selecting a first k item attributes having the highest selection priority from the plurality of item attributes, k being a positive integer.

[0240] In some embodiments, the distribution information of the first item attribute comprises at least one of: a total frequency, a first frequency, a second frequency, a third frequency, and a fourth frequency, wherein: the total frequency is a frequency of the first item attribute appearing in the recommended items of each style under the first category; the first frequency is a frequency of attribute content of the first item attribute being the same in the same-style recommended items under the first category; the second frequency is a frequency of attribute content of the first item attribute being different in the same-style recommended items under the first category; the third frequency is a frequency of attribute content of the first item attribute being the same in the non-same-style recommended items under the first category; and the fourth frequency is a frequency of attribute content of the first item attribute being different in the non-same-style recommended items under the first category.

[0241] In some embodiments, the attribute selection unit is configured to: determine a first weight according to the total frequency; determine a second weight according to the first frequency and the fourth frequency, the influence of the total frequency on the first weight being greater than the influence of the first frequency and the fourth frequency on the second weight; determine a third weight according to the second frequency and the third frequency, the influence of the first frequency and the fourth frequency on the second weight being greater than the influence of the second frequency and the third frequency on the third weight; and determine a selection priority of the first item attribute according to the first weight, the second weight, and the third weight.

[0242] In some embodiments, the category attribute rule comprises attribute indication information, the attribute indication information being used to indicate at least one key attribute for distinguishing styles of recommended items under the first category; and the identification determination module 720 is configured to: determine the at least one key attribute from the plurality of item attributes according to the attribute indication information; and determine the at least one key attribute as the identification information of the first recommended item.

[0243] In some embodiments, the result determination module 730 comprises: an input determination unit configured to determine input information of a style discrimination model according to the identification information of the first recommended item and the identification information of the second recommended item, the style discrimination model being obtained by training training data based on a category attribute system, the category attribute system being used to classify recommended items in a recommendation platform; and a result acquisition unit configured to obtain the style discrimination result by processing the input information through the style discrimination model.

[0244] In some embodiments, the identification information of the first recommended item includes at least one item attribute of the plurality of item attributes, and the at least one item attribute has a selection priority; the input determination unit is configured to sort the at least one item attribute included in the identification information of the first recommended item according to the selection priority to obtain first input sub-information; if a character length of the first input sub-information is greater than a length threshold, truncate i item attributes from an end of the first input sub-information, where i is a positive integer; sort the at least one item attribute included in the identification information of the second recommended item according to the selection priority to obtain second input sub-information; if a character length of the second input sub-information is greater than the length threshold, truncate j item attributes from an end of the second input sub-information, where j is a positive integer; and concatenate the first input sub-information and the second input sub-information to obtain the input information of the style discrimination model.

[0245] In some embodiments, the identification information of the first recommended item includes at least one key attribute for representing a style of a recommended item; and the result determination module 730 is configured to determine that the style discrimination result is used to represent that the first recommended item and the second recommended item are same-item recommended items, in a case where the first recommended item and the second recommended item have consistent attribute contents in at least one key attribute.

[0246] In some embodiments, the information acquisition module 710 is configured to: acquire recommended item information of the first recommended item from a recommendation platform, the recommendation platform being configured to provide recommended items to a user; and identify the recommended item information of the first recommended item by using an entity recognition model to obtain attribute information of the first recommended item, the entity recognition model being obtained by adjusting training data, and the training data being obtained by extracting recommended item information in at least one recommendation platform based on a category attribute system.

[0247] It should be noted that the apparatus provided in the above embodiments, when realizing its functions, is only exemplified by the above division of functional modules, and in actual applications, the above functions can be completed by different functional modules according to needs, that is, the content structure of the device is divided into different functional modules to complete all or part of the above-described functions. In addition, the apparatus and method embodiments provided in the above embodiments belong to the same concept, and the specific implementation process is detailed in the method embodiments, which will not be described here. The beneficial effects of the apparatus provided in the above embodiments are described in the description of the method embodiments, which will not be described here.

[0248] Figure 8 A structural block diagram of a computer device provided in an example embodiment of the present application is shown. The recommended item style discrimination device 800 can be the computer device or the server introduced above.

[0249] Generally, the computer device 800 includes a processor 801 and a memory 802.

[0250] The processor 801 can include one or more processing cores, such as a 4-core processor, an 8-core processor, and the like. The processor 801 can be implemented in the form of at least one of a DSP (Digital Signal Processing), an FPGA (Field Programmable Gate Array), a PLA (Programmable Logic Array). The processor 801 can also include a main processor and a coprocessor. The main processor is a processor for processing data in an awake state, also referred to as a CPU (Central Processing Unit). The coprocessor is a low-power processor for processing data in a standby state. In some embodiments, the processor 801 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content to be displayed on the display screen. In some embodiments, the processor 801 can further include an AI processor for processing machine learning-related computing operations.

[0251] The memory 802 can include one or more computer-readable storage media, which can be tangible and non-transitory. The memory 802 can also include a high-speed random access memory, and a non-volatile memory such as one or more disk storage devices, flash storage devices. In some embodiments, the non-transitory computer-readable storage medium in the memory 802 stores at least one program loaded and executed by the processor 801 to implement the recommended item style discrimination method provided by the above method embodiments.

[0252] The embodiments of the present application also provide a computer-readable storage medium, which stores a computer program. The computer program is loaded and executed by the processor to implement the recommended item style discrimination method provided by the above method embodiments.

[0253] The computer readable media can include computer storage media and communication media. Computer storage media includes volatile and non-volatile, removable and non-removable media implemented in any method or technology for storage of information such as computer readable instructions, data structures, program modules or other data. Computer storage media includes, but is not limited to, RAM, ROM, EPROM, EEPROM, flash memory or other solid state memory technology, CD-ROM, DVD, or other optical disk storage, magnetic cassettes, magnetic tape, magnetic disk storage or other magnetic storage devices. It should be understood that computer storage media does not limit the computer readable media to the foregoing examples. However, computer storage media excludes modulated data signals and carrier waves.

[0254] The embodiment of the present application further provides a computer program product, which comprises a computer program stored in a computer readable storage medium, and a processor reads and executes the computer program from the computer readable storage medium to realize the recommendation item style discrimination method provided by each method embodiment.

[0255] It should be understood that "multiple" mentioned herein refers to two or more. The "and / or" describes the association relationship of the associated objects, which means that there can be three relationships, for example, A and / or B can represent the following three cases: A exists alone, A and B exist together, and B exists alone. The character " / " generally represents that the associated objects before and after it are in an "or" relationship.

[0256] The above only describes optional embodiments of the present application and is not intended to limit the present application. Any modification, equivalent switching, improvement, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for identifying the style of recommended items, characterized in that: The method comprises: Acquire attribute information of a first recommended item, where the attribute information of the first recommended item includes multiple item attributes of the first recommended item; Determining identification information of the first recommendation according to a category attribute rule and attribute information of the first recommendation, wherein the category attribute rule is a rule for determining identical recommendations within the first category to which the first recommendation belongs, and the identification information of the first recommendation is used to characterize the first recommendation within the first category; A style discrimination result is determined based on the identification information of the first recommendation and the identification information of the second recommendation. The identification information of the second recommendation is determined in the same manner as the identification information of the first recommendation. The style discrimination result is used to characterize the style similarity between the first recommendation and the second recommendation.

2. The method according to claim 1, characterized in that The category attribute rules include attribute distribution information, which is used to represent the distribution of the item attributes among the recommended items of each style under the first category; The determining, based on the category attribute rule and the attribute information of the first recommendation, identification information of the first recommendation includes: selecting at least one item attribute from the plurality of item attributes according to the attribute distribution information; The recommendation title of the first recommendation and the at least one item attribute are concatenated to obtain identification information of the first recommendation.

3. The method according to claim 2, characterized in that The selecting at least one item attribute from the plurality of item attributes according to the attribute distribution information includes: For a first item attribute among the multiple item attributes, determining a selection priority for the first item attribute based on distribution information of the first item attribute in the attribute distribution information, the selection priority of the first item attribute being used to indicate a priority for selecting the first item attribute from the multiple item attributes, the distribution information of the first item attribute being used to indicate a distribution of the first item attribute among multiple recommended items under the first category, the multiple recommended items being from at least one recommendation platform, the recommendation platform being used to provide recommendations to the user; Selecting the top k item attributes with the highest selection priority from the multiple item attributes to obtain the at least one item attribute, where k is a positive integer.

4. The method according to claim 3, characterized in that The distribution information of the first item attribute includes at least one of the following: total frequency, first frequency, second frequency, third frequency, and fourth frequency, wherein: The total frequency is the frequency at which the first item attribute appears in the recommended items of each style under the first category; The first frequency is the frequency at which the attribute content of the first item attribute is the same among the same recommended items under the first category; The second frequency is the frequency at which the attribute content of the first item attribute is different among the same recommended items under the first category; The third frequency is the frequency at which the attribute content of the first item attribute is the same among the non-identical recommended items under the first category; The fourth frequency is the frequency at which the attribute content of the first item attribute is different among non-identical recommended items under the first category.

5. The method according to claim 4, characterized in that The determining, based on distribution information of the first item attribute in the attribute distribution information, a selection priority of the first item attribute includes: determining a first weight according to the total frequency; determining a second weight according to the first frequency and the fourth frequency, wherein an influence of the total frequency on the first weight is greater than an influence of the first frequency and the fourth frequency on the second weight; determining a third weight according to the second frequency and the third frequency, wherein an influence of the first frequency and the fourth frequency on the second weight is greater than an influence of the second frequency and the third frequency on the third weight; A selection priority of the first item attribute is determined according to the first weight, the second weight, and the third weight.

6. The method according to claim 1, characterized in that The category attribute rule includes attribute indication information, where the attribute indication information is used to indicate at least one key attribute for distinguishing recommended item styles under the first category; The determining, based on the category attribute rule and the attribute information of the first recommendation, identification information of the first recommendation includes: determining the at least one key attribute from the plurality of item attributes according to the attribute indication information; The at least one key attribute is determined as identification information of the first recommendation.

7. The method according to claim 1, characterized in that The determining of a style identification result according to the identification information of the first recommendation item and the identification information of the second recommendation item includes: Determining input information for a style discrimination model based on the identification information of the first recommended item and the identification information of the second recommended item, wherein the style discrimination model is trained based on first training data constructed based on a category attribute system used to classify recommended items in the recommendation platform; The input information is processed by the style discrimination model to obtain the style discrimination result.

8. The method according to claim 7, characterized in that The identification information of the first recommendation includes at least one item attribute among the multiple item attributes, and the at least one item attribute has a selection priority; The step of determining input information of a style discrimination model according to the identification information of the first recommendation item and the identification information of the second recommendation item includes: sorting at least one item attribute included in the identification information of the first recommendation according to the selection priority to obtain first input sub-information; If the character length of the first input sub-information is greater than the length threshold, truncating i item attributes from the end of the first input sub-information, where i is a positive integer; sorting at least one item attribute included in the identification information of the second recommendation according to the selection priority to obtain second input sub-information; If the character length of the second input sub-information is greater than the length threshold, truncating j item attributes from the end of the second input sub-information, where j is a positive integer; The first input sub-information and the second input sub-information are concatenated to obtain input information of the style discrimination model.

9. The method according to claim 1, characterized in that The identification information of the first recommended item includes at least one key attribute used to characterize the style of the recommended item; The determining of a style identification result according to the identification information of the first recommendation item and the identification information of the second recommendation item includes: If the first recommended item and the second recommended item have the same attribute content in at least one of the key attributes, the style identification result is determined to indicate that the first recommended item and the second recommended item are the same recommended item.

10. The method according to claim 1, characterized in that The obtaining of attribute information of the first recommendation includes: obtaining recommendation information of the first recommendation from a recommendation platform, wherein the recommendation platform is used to provide recommendations to users; The recommended item information of the first recommended item is identified through an entity recognition model to obtain attribute information of the first recommended item. The entity recognition model is trained by second training data, and the second training data is extracted from the recommended item information in at least one of the recommendation platforms based on a category attribute system.

11. A device for identifying the style of recommended items, characterized in that: The device comprises: an information acquisition module, configured to acquire attribute information of a first recommended item, wherein the attribute information of the first recommended item includes a plurality of item attributes of the first recommended item; an identification determination module, configured to determine identification information of the first recommendation item based on a category attribute rule and attribute information of the first recommendation item, wherein the category attribute rule is a rule for determining similar recommendations within the first category to which the first recommendation item belongs, and the identification information of the first recommendation item is used to characterize the first recommendation item within the first category; A result determination module is used to determine a style discrimination result based on the identification information of the first recommendation object and the identification information of the second recommendation object. The identification information of the second recommendation object is determined in the same manner as the identification information of the first recommendation object. The style discrimination result is used to characterize the style similarity between the first recommendation object and the second recommendation object.

12. A computer device, characterized in that: The computer device includes a processor and a memory, wherein a computer program is stored in the memory, and the computer program is loaded and executed by the processor to implement the method for determining the style of recommended items according to any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that The storage medium stores a computer program, which is loaded and executed by a processor to implement the method for determining the style of recommended items according to any one of claims 1 to 10.

14. A computer program product, characterized in that The computer program product includes a computer program, which is stored in a computer-readable storage medium. A processor reads and executes the computer program from the computer-readable storage medium to implement the method for determining the style of recommended items according to any one of claims 1 to 10.

Images