Method and device for determining interest tag of object and electronic equipment

Abstract

The embodiment of the invention provides a method and device for determining an interest tag of an object, electronic equipment and a computer readable storage medium, and relates to the field of data processing. The method comprises the steps of determining an object portrait of a target object, a to-be-identified tag set and a first historical interest tag set; and inputting the object portrait of the target object, the to-be-identified tag set and the first historical interest tag set into a pre-trained interest tag identification model to obtain an identification result set output by the interest tag identification model, the identification result set including an identification result indicating whether each to-be-identified tag is an interest tag of the target object. The embodiment of the invention is used for accurately determining the interest tag of the target object in the plurality of to-be-identified tags and performing targeted commodity recommendation for the target object.

Description

Technical Field

[0001] This application relates to the technical field of computers and data processing. Specifically, this application relates to a method, apparatus, electronic device, and computer-readable storage medium for determining interest tags of an object. Background Art

[0002] In a recommendation system, a large number of object interest tags, that is, objects' truly interested products, are required to recommend products to an object. According to the behavior information of the object, a lot of tags related to the object can be collected. These tags include the object's interest tags and non-interest tags. Therefore, it is necessary to process the object's tags, delete the object's non-interest tags, and then make a reasonable product recommendation for the object.

[0003] Tags related to an object can be obtained from multiple data sources. For example, tags of the object can be obtained from the advertisements browsed by the object and the articles read by the object. In the prior art, the object tags obtained from different data sources are analyzed, and the tags that appear in different data sources are used as the object's interest tags. However, the number of object interest tags obtained by this method is very small. Based on these few interest tags, it is difficult to accurately analyze the object's true preferences, so accurate recommended content cannot be provided for the object. Summary of the Invention

[0004] Embodiments of this application provide a method, apparatus, electronic device, computer-readable storage medium, and computer program product for determining interest tags of an object, which can solve the above problems of the prior art. The technical solutions are as follows:

[0005] According to the first aspect of the embodiments of this application, a method for determining interest tags of an object is provided. The method includes:

[0006] Determine the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object;

[0007] Input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model, and obtain a recognition result set output by the interest tag recognition model. The recognition result set includes the recognition results of whether each tag to be recognized is an interest tag of the target object;

[0008] Wherein, the interest tag recognition model is trained with the object portraits, the first sample tag sets, and the second historical interest tag sets of multiple first sample objects as training samples, and the exposure behaviors of each first sample tag in the first sample tag sets of the corresponding first sample objects as training labels;

[0009] Among them, the types of the first sample tags include common sense interest tags and counter - common sense interest tags related to portrait - attribute pairs;

[0010] The common sense interest tag refers to: for a sample object group including the corresponding same portrait - attribute pair, a tag with a relatively high probability that the exposure behavior is a click behavior;

[0011] The counter - common sense interest tag refers to: for a sample object group including the corresponding same portrait - attribute pair, a tag with a relatively low probability that the exposure behavior is a click behavior.

[0012] According to the second aspect of the embodiments of the present application, there is provided a device for determining an interest tag of an object. The device includes:

[0013] An obtaining module, configured to determine an object portrait of a target object, a set of tags to be recognized, and a first historical interest tag set;

[0014] An input module: configured to input the object portrait of the target object, the set of tags to be recognized, and the first historical interest tag set into a pre - trained interest tag recognition model, and obtain a recognition result set output by the interest tag recognition model. The recognition result set includes recognition results indicating whether each tag to be recognized is an interest tag of the target object;

[0015] Among them, the interest tag recognition model is trained with the object portraits of multiple first sample objects, a first sample tag set, and a second historical interest tag set as training samples, and the exposure behaviors of each first sample tag in the first sample tag set of the corresponding first sample object as training labels;

[0016] Among them, the types of the first sample tags include common sense interest tags and counter - common sense interest tags related to portrait - attribute pairs;

[0017] The common sense interest tag refers to: for a sample object group including the corresponding same portrait - attribute pair, a tag with a relatively high probability that the exposure behavior is a click behavior;

[0018] The counter - common sense interest tag refers to: for a sample object group including the corresponding same portrait - attribute pair, a tag with a relatively low probability that the exposure behavior is a click behavior.

[0019] According to the third aspect of the embodiments of the present application, there is provided an electronic device. The electronic device includes a memory, a processor, and a computer program stored on the memory. The processor executes the computer program to implement the steps of the method in the first aspect.

[0020] According to the fourth aspect of the embodiments of the present application, there is provided a computer - readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of the method in the first aspect are implemented.

[0021] According to a fifth aspect of the embodiments of the present application, there is provided a computer program product including a computer program which, when executed by a processor, implements the steps of the method in the first aspect.

[0022] The beneficial effects brought by the technical solutions provided in the embodiments of the present application are as follows:

[0023] By inputting the object portrait of the target object, the set of tags to be recognized, and the first historical interest tag set into a pre-trained interest tag recognition model, a recognition result set output by the interest tag recognition model is obtained. Since the interest tag recognition model uses the object portraits of multiple first sample objects, the first sample tag set, and the second historical interest tag set as training samples, and uses the exposure behavior of each first sample tag in the first sample tag set of the corresponding sample object as a training tag, the types of the first sample tags include common sense interest tags, and the common sense interest tags are related to portrait attribute pairs, that is, the first sample tags of different portrait attribute pairs are different. The present application statistically analyzes the click behavior of the sample object group at the granularity of portrait attribute pairs to obtain common sense interest tags. By analyzing the object portrait of the sample object, the common sense interest tags, the corresponding exposure behavior, and the second historical interest tag set, the model can analyze the interest tags in each common sense interest tag of the sample object. Then, after the model training is completed, it can accurately determine, based on the object portrait of the target object, the set of tags to be recognized, and the first historical interest tag set, the recognition result of whether the tag to be recognized is an interest tag of the target object, that is, it can determine the recognition result of whether each tag to be recognized generated by multiple data sources is an interest tag of the target object. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the following will briefly introduce the drawings required for description in the embodiments of the present application.

[0025] Figure 1 It is a schematic diagram of the system architecture for implementing the method for determining the interest tags of an object provided in the embodiments of the present application;

[0026] Figure 2 It is a schematic flowchart of a method for determining the interest tags of an object provided in the embodiments of the present application;

[0027] Figure 3 It is a schematic flowchart of another method for determining the interest tags of an object provided in the embodiments of the present application;

[0028] Figure 4 It is a schematic flowchart of the iterative training process of an interest tag recognition model provided in the embodiments of the present application;

[0029] Figure 5 Schematic flowchart of the process for determining common sense interest tags and counter common sense interest tags provided by an embodiment of the present application;

[0030] Figure 6 Schematic flowchart of the process for determining the first sample tag set provided by an embodiment of the present application;

[0031] Figure 7 Schematic flowchart of the process for determining the interest tags of another object provided by an embodiment of the present application;

[0032] Figure 8 Schematic diagram of the application scenario of the method for determining the interest tags of an object provided by an embodiment of the present application

[0033] Figure 9 Schematic diagram of the structure of a device for determining the interest tags of an object provided by an embodiment of the present application;

[0034] Figure 10 Schematic diagram of the structure of an electronic device for determining the interest tags of an object provided by an embodiment of the present application. Detailed implementation manners

[0035] The embodiments of the present application will be described below with reference to the accompanying drawings in the present application. It should be understood that the embodiments described below in conjunction with the accompanying drawings are exemplary descriptions for explaining the technical solutions of the embodiments of the present application, and do not constitute limitations on the technical solutions of the embodiments of the present application.

[0036] Those skilled in the art of the present technology can understand that unless specifically stated otherwise, the singular forms "a", "an", "the" and "said" used herein may also include the plural forms. It should be further understood that the terms "including" and "comprising" used in the embodiments of the present application mean that the corresponding features can be implemented as the presented features, information, data, steps, operations, elements, and / or components, but do not exclude the implementation of other features, information, data, steps, operations, elements, components, and / or their combinations supported by the art of the present technology. It should be understood that when we say an element is "connected" or "coupled" to another element, this element can be directly connected or coupled to the other element, or it can mean that this element and the other element establish a connection relationship through an intermediate element. In addition, the "connection" or "coupling" used herein may include a wireless connection or a wireless coupling. The term "and / or" used herein indicates at least one of the items defined by this term. For example, "A and / or B" can be implemented as "A", or implemented as "B", or implemented as "A and B".

[0037] In the embodiments of the present application, the term "module" or "unit" refers to a computer program with a predetermined function or a part of a computer program, which works together with other related parts to achieve a predetermined goal, and can be fully or partially implemented by using software, hardware (such as a processing circuit or a memory), or a combination thereof. Similarly, a processor (or multiple processors or memories) can be used to implement one or more modules or units. In addition, each module or unit can be a part of an overall module or unit that includes the functions of that module or unit.

[0038] To make the objectives, technical solutions, and advantages of the present application clearer, the embodiments of the present application will be further described in detail below with reference to the accompanying drawings.

[0039] First, several terms related to the present application will be introduced and explained:

[0040] Label: That is, a commodity. Clicking on a label in the present application means clicking on or browsing an advertisement, article, etc. related to that label.

[0041] Interest label: A commodity that the object is truly interested in.

[0042] Historical interest label: Based on the behavior information of the object within a certain period in the past, the commodity that the object is truly interested in within a certain period is determined.

[0043] Object portrait: The personal information of the object, such as age, gender, place of residence, education level, type of work, etc.

[0044] The 80 / 20 principle: The 80 / 20 principle is also known as the Pareto principle, which was discovered by the Italian economist Pareto in the late 19th century and early 20th century. This principle holds that in any set of data, the most important only accounts for a small part, about 20%, and the remaining 80%, although in the majority, is secondary. Therefore, it is also called the 80 / 20 principle. The 80 / 20 principle applies to most industries and can solve problems such as time management problems, key customer problems, wealth distribution problems, resource allocation problems, core product problems, key talent problems, core profit problems, etc.

[0045] The Sigmoid function is also called the Logistic function, which is used for the output of hidden layer neurons and has a value range of (0, 1). It can map a real number to the interval (0, 1) and can be used for binary classification. It has good effects when the features are relatively complex or not very different. The Sigmoid function is an activation function in neural networks and is a smooth and strictly monotonic saturation function

[0046] Cloud technology is a hosting technology that unifies a series of resources such as hardware, software, and networks within a wide area network or a local area network to achieve data calculation, storage, processing, and sharing.

[0047] Cloud technology refers to the general term for network technology, information technology, integration technology, management platform technology, application technology, etc. applied based on the cloud computing business model. It can form a resource pool, be used on demand, and is flexible and convenient. Cloud computing technology will become an important support. The background services of the technical network system require a large amount of computing and storage resources, such as video websites, picture websites, and more portal websites. With the highly developed and applied Internet industry, in the future, each item may have its own identification mark and needs to be transmitted to the background system for logical processing. Data at different levels will be processed separately, and various industry data requires a powerful system support, which can only be achieved through cloud computing.

[0048] Big data refers to a collection of data that cannot be captured, managed, and processed by conventional software tools within a certain time range. It is a massive, high-growth, and diverse information asset that requires a new processing mode to have stronger decision-making power, insight discovery ability, and process optimization ability. With the advent of the cloud era, big data has also attracted more and more attention. Big data requires special technologies to effectively process a large amount of data tolerated over time. Technologies applicable to big data include massively parallel processing databases, data mining, distributed file systems, distributed databases, cloud computing platforms, the Internet, and scalable storage systems.

[0049] Natural Language Processing (NLP) is an important direction in the fields of computer science and artificial intelligence. It studies various theories and methods that can achieve effective communication between humans and computers in natural language. Natural language processing involves natural language, that is, the language people use in daily life, and is closely related to linguistics research; at the same time, it involves computer science and mathematics. It is a science that integrates linguistics, computer science, and mathematics. Therefore, the research in this field will involve natural language, that is, the language people use in daily life, so it has a close connection with linguistics research. The pre-trained model, an important technology for model training in the field of artificial intelligence, is developed from the large language model in the NLP field. After fine-tuning, the large language model can be widely applied to downstream tasks. Natural language processing technologies usually include text processing, semantic understanding, machine translation, robot question answering, knowledge graphs, and other technologies.

[0050] The method, device, electronic device, computer-readable storage medium, and computer program product for determining the interest tags of the object provided in this application aim to solve the above technical problems of the prior art.

[0051] The technical solutions of the embodiments of the present application and the technical effects brought about by the technical solutions of the present application will be described below through the description of several exemplary embodiments. It should be noted that the following embodiments can be referred to, learned from, or combined with each other. For the same terms, similar features, and similar implementation steps in different embodiments, they will not be described repeatedly.

[0052] Figure 1 FIG. is a schematic diagram of the system architecture for the method of determining the interest tags of an object provided in an embodiment of the present application, which includes a terminal 101 and a server 102. The terminal 101 and the server 102 can be directly or indirectly connected through wired or wireless communication means, and the embodiments of the present application do not limit this.

[0053] In some embodiments, the terminal 101 is a smart phone, a tablet computer, a laptop computer, a desktop computer, a smart speaker, a smart watch, a smart voice interaction device, a smart home appliance, a vehicle-mounted terminal, etc., but is not limited thereto. The terminal 101 is installed with an application program that supports the determination of the interest tags of an object. The application program can be a social application program, a game application, or a multimedia application, and the embodiments of the present application do not limit this.

[0054] Those skilled in the art can know that the number of the above terminals can be more or less. For example, the above terminal can be only one, or the above terminals can be dozens or hundreds, or more. The embodiments of the present application do not limit the number and device type of the terminals.

[0055] In some embodiments, the server 102 is an independent physical server, and can also be a server cluster or a distributed system composed of multiple physical servers, and can also be a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, Content Delivery Network (CDN), and big data and artificial intelligence platforms. The server 102 is used to provide background services for the application program that supports image acquisition. In some embodiments, the server 102 undertakes the main computing work, and the terminal 101 undertakes the secondary computing work; or, the server 102 undertakes the secondary computing work, and the terminal 101 undertakes the main computing work; or, a distributed computing architecture is adopted between the server 102 and the terminal 101 for collaborative computing.

[0056] For example, the method for determining the interest tags of an object in the embodiments of the present application can be completed by the terminal. The following will describe the solution in the embodiments of the present application with the terminal as the execution subject.

[0057] An embodiment of the present application provides a method for determining the interest tags of an object, asFigure 2 As shown in the figure, the method includes S101 and S102.

[0058] S101. Determine the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object.

[0059] In order to obtain the interest tags of the target object (i.e., the tags that the target object is interested in), embodiments of the present application obtain the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object.

[0060] In embodiments of the present application, the object portrait of an object is the personal information of the object, including portrait attributes in various dimensions, such as at least one of age, gender, education level, work type, place of residence, and work type.

[0061] It should be noted that, when collecting and processing relevant data in various embodiments of the present application, in practical applications, the informed consent or separate consent of the personal information subject should be obtained in strict accordance with the requirements of relevant national laws and regulations, and subsequent data use and processing behaviors should be carried out within the scope authorized by laws and regulations and the personal information subject.

[0062] The set of tags to be recognized of an object includes multiple tags to be recognized of the object. A tag to be recognized is a tag whose interest tag of the user has not been determined, that is, the tag to be recognized in embodiments of the present application.

[0063] The first historical interest tag set of an object includes multiple first historical interest tags of the object. The first historical interest tag is an interest tag of the object within a preset time period. Embodiments of the present application do not limit the specific duration of the preset time period, such as six months. In some embodiments, the first historical interest tag can be an interest tag obtained from multiple data sources. For example, the advertisements clicked by the object are obtained from an advertisement data source, the articles clicked by the object are obtained from a reading data source, and the transaction data in an advertisement exchange (ADX).

[0064] In embodiments of the present application, the first historical interest tag set of the target object may include identity identification information and may be pre-stored in an interest tag recognition database. By determining the identity identification of the target object, the first historical interest tag set corresponding to the identity identification is automatically extracted from the interest tag recognition database.

[0065] S102. Input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain a recognition result set output by the interest tag recognition model. The recognition result set includes the recognition results of whether each tag to be recognized is an interest tag of the target object.

[0066] In the embodiment of the present application, the object portrait of the target object determined in step S101, the set of tags to be recognized, and the first historical interest tag set are combined into a multi-dimensional data and input into the interest tag recognition model. For example, an instance of an input data is <42312; male, from Hubei, 30 years old, undergraduate degree, private enterprise; football game, lipstick; World Cup, FIFA, football, running>, where 42312 is the object identifier of the target object, male, from Hubei, 30 years old, undergraduate degree, and private enterprise are the respective portrait attributes of the object portrait, football game, lipstick are the tags to be recognized, and World Cup, FIFA, football, and running are the first historical interest tag set.

[0067] The interest tag recognition model in the embodiment of the present application can output the recognition results of the tags to be recognized in the set of tags to be recognized, and the recognition results of all the tags to be recognized constitute a recognition result set. The output form can be to output whether the tag to be recognized is an interest tag of the target object, or to only output the tags to be recognized in the set of tags to be recognized that are the interest tags of the target object. Taking the above embodiment as an example, when the output form is to output whether the tag to be recognized is an interest tag of the target object, the output can be <Football is an interest tag, Lipstick is not an interest tag>, and when the output form is to only output the tags to be recognized in the set of tags to be recognized that are the interest tags of the target object, the output can be <Football is an interest tag>.

[0068] In the embodiment of the present application, the training samples of the interest tag recognition model are the object portraits of multiple first sample objects, the first sample tag set, and the second historical interest tag set, and the training tags are the exposure behaviors of each first sample tag in the first sample tag set of the corresponding first sample object.

[0069] In the embodiment of the present application, the type of each first sample tag of the first sample object is a common sense interest tag. Each first sample tag is defined as being related to the object attribute pair of the common sense interest tag, that is, for the sample object groups of different object attribute pairs, the common sense interest tags may be different. For example, the common sense interest tags of the sample object group of "30-year-old + female" may be lipstick and dress, while the common sense interest tags of the sample object group of "50-year-old + female" may be tissue and thermos cup.

[0070] A common sense interest tag related to an object attribute pair means: among the sample groups containing this portrait attribute pair, the tag has a relatively high probability of being clicked or browsed after being exposed. For example, the click behaviors of different tags of the sample group containing the object attribute pair (age = 30, gender = male) are counted to obtain one hundred tags. According to the click-through rate of each tag, each tag is sorted from largest to smallest, and the top 80 tags are used as the common sense interest tags.

[0071] It should be noted that the sample object group can refer to the above-mentioned multiple first sample objects, or other sample objects, which is not limited in this application.

[0072] In the embodiment of the present application, the second historical interest tag set of the first sample object includes the second historical interest tags of the first sample object. The second historical interest tag is the interest tag of the first sample object within a preset time period. The preset time period can be the first six months during model training, and the second historical interest tag can be an interest tag obtained from multiple data sources. For example, the advertisements clicked by the object are obtained from the advertisement data source, the articles clicked by the object are obtained from the reading data source, and the transaction data in the ADX.

[0073] By inputting the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model, a recognition result set output by the interest tag recognition model is obtained. Since the interest tag recognition model uses the object portraits, the first sample tag sets, and the second historical interest tag sets of multiple first sample objects as training samples, and uses the exposure behaviors of each first sample tag in the first sample tag set of the corresponding sample object as training tags, the types of the first sample tags include common sense interest tags, and the common sense interest tags are related to the portrait attribute pairs. That is, the first sample tags of different portrait attribute pairs are different. In this application, the click behaviors of the sample object group are statistically analyzed at the granularity of the portrait attribute pairs to obtain common sense interest tags. The model can analyze the interest tags in each common sense interest tag of the sample object by analyzing the object portrait, the common sense interest tags, the corresponding exposure behaviors, and the second historical interest tag set of the sample object. Furthermore, after the model training is completed, it is possible to accurately determine, based on the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object, the recognition result of whether the tag to be recognized is an interest tag of the target object, that is, it is possible to determine the recognition result of whether each tag to be recognized generated by multiple data sources is an interest tag of the target object.

[0074] Based on the above embodiments, as an optional embodiment, inputting the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain the recognition result set output by the interest tag recognition model includes:

[0075] S201. Input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain vector representations of each first historical interest tag and the object portrait of the target object, and the semantic similarity between each tag to be recognized and each first historical interest tag;

[0076] S202. For each label to be recognized, based on the semantic similarity between the label to be recognized and each first historical interest label, perform weighted summation on the vector representations of each first historical interest label to obtain the first vector of the label to be recognized.

[0077] S203. For each label to be recognized, combine the first vector of the label to be recognized and the vector representation of the object portrait of the target object to obtain the second vector of the label to be recognized, where the second vector is used to characterize the degree of interest of the target object in the label to be recognized.

[0078] S204. For each label to be recognized, obtain the recognition result of the label to be recognized output by the interest label recognition model based on the second vector of the label to be recognized.

[0079] Please refer to Figure 3 , which exemplarily shows a flowchart of another method for determining the interest label of an object provided by an embodiment of the present application.

[0080] In the embodiment of the present application, after inputting the object portrait of the target object, the set of labels to be recognized, and the set of first historical interest labels into the interest label recognition model, the vector representations of the object portrait of the target object, each label to be recognized, and each first historical interest label are obtained.

[0081] Meanwhile, for each label to be recognized in the set of labels to be recognized, determine the semantic similarity between the label to be recognized and each first historical interest label.

[0082] In the embodiment of the present application, the semantic similarity between the label to be recognized and each first historical interest label can be determined according to the inner product of the vector representation of the label to be recognized and the vector representation of each first historical interest label. In addition, an attention mechanism module can be added to the emotion label recognition model, and the correlation score between the label to be recognized and each first historical interest label can be obtained based on the vector representation of the label to be recognized and the vector representations of each first historical interest label, so as to determine the semantic similarity between the label to be recognized and each first historical interest label.

[0083] In some embodiments, the semantic similarity between a label to be recognized and each first historical interest label can also be obtained before inputting into the interest label recognition model, and the acquisition method is not limited in the present application. For example, a deep learning model (such as a convolutional neural network, a recurrent neural network, etc.) is used to encode the label to be recognized and each first historical interest label, then the similarity between the two encoded vectors is compared, and finally the semantic similarity between the label to be recognized and each first historical interest label is obtained.

[0084] In the embodiment of the present application, the semantic similarity between a tag to be recognized and each first historical interest tag is used as the weight of the corresponding first historical interest tag for the tag to be recognized. Based on the vector representation of each first historical interest tag, a weighted sum is performed on each first historical interest tag, and the obtained vector representation is used as the first vector of the tag to be recognized. That is to say, in the embodiment of the present application, through the semantic similarity between the tag to be recognized and each first historical interest tag, the first vector of the tag to be recognized is defined as the vector obtained by the weighted sum of the vector representations of each first historical interest tag. For example, the tag to be recognized is tag 1, the first historical interest tags are tag 2, tag 3, and tag 4 respectively, the vector representation of tag 2 is B, the vector representation of tag 3 is C, the vector representation of tag 4 is D, and the semantic similarities between tag 1 and tag 2, tag 1 and tag 3, tag 1 and tag 4 are 0.5, 0.2, and 0.4 respectively. Then the first vector of tag 1 is: 0.5B + 0.2C + 0.4D.

[0085] In the embodiment of the present application, the vector representation of the object portrait of the target object obtained in step S201 and the first vector of each tag set to be recognized in the tag set to be recognized of the target object obtained in step S202 are combined to obtain the second vector of each tag to be recognized in the tag set to be recognized. The combination method can be that for the first vector representation of a tag to be recognized, it is concatenated with the vector representation of the object portrait of the target object. For example, the first vector representation of the tag to be recognized is concatenated to the tail of the vector representation of the object portrait of the target object. For example, the first vector of the tag to be recognized is <a1, a2, a3,..., am>, and the vector representation of the object portrait is <b1, b2, b3,..., bn>. Then the second vector of the tag to be recognized is <b1, b2, b3,..., bn, a1, a2, a3,..., am>. The specific concatenation method and the specific combination method are not limited in the embodiment of the present application.

[0086] In the embodiment of the present application, by combining the first vector representation of each tag to be recognized of the target object with the vector representation of the object portrait of the target object, the second vector of each tag to be recognized of the target object is obtained, which is used to characterize the degree of interest of the target object in each tag to be recognized.

[0087] In the embodiment of the present application, for each tag to be recognized of the target object, the second vector representation of each tag to be recognized is obtained. According to the second vector representation of each tag to be recognized, the recognition result of the corresponding tag to be recognized is obtained as the output result of the interest tag recognition model.

[0088] Such as Figure 3As shown in the figure, in the embodiment of the present application, the way to obtain the recognition result of each tag to be recognized may be to input the second vector of each tag to be recognized into the sigmoid function. The function value output by the sigmoid function is the probability value that the corresponding tag to be recognized is an interest tag. Compare this probability value with a preset threshold. If the probability value is greater than the preset threshold, the tag to be recognized corresponding to this probability value is determined as the interest tag of the target object; otherwise, the tag to be recognized corresponding to this probability value is not determined as the interest tag of the target object.

[0089] It should be noted that Figure 3 This shows an output method provided by the embodiment of the present application, that is, output whether the tag to be recognized is an interest tag or the tag to be recognized is not an interest tag. As another optional embodiment, the output method may be to only output the tags to be recognized in the tag set to be recognized that are interest tags, and not output the tags to be recognized that are not interest tags.

[0090] In the embodiment of the present application, by using the first historical interest tags of the target object as the reference data of the target object, the first vector of the tag to be recognized is obtained through the semantic similarity between the tag to be recognized of the target object and each first historical interest tag, and the second vector of the tag to be recognized is obtained by combining the vector representation of the object portrait of the target user. And the interest tag recognition model in the embodiment of the present application is trained according to the object portrait of the first sample object and other data. That is to say, in the embodiment of the present application, the degree of interest of the target object in the tag to be recognized is comprehensively determined by integrating the object portrait of the target object, the first historical interest tags, as well as the object portrait of the first sample object, the first sample interest tags and the second historical interest tags and other factors. Therefore, the method for determining the interest tags of the object provided by the embodiment of the present application has higher accuracy than other methods. At the same time, for each tag to be recognized obtained from multiple data sources, it is determined whether each tag to be recognized is an interest tag of the target object.

[0091] Based on the object portraits, the first sample tag sets and the second historical interest tag sets of multiple first sample objects as training samples, and using the exposure behaviors of each first sample tag in the first sample tag set of the corresponding first sample object as training labels, perform multiple rounds of iterative training on the initial model until the iterative stop condition is met;

[0092] Wherein, each round of iterative training includes:

[0093]

[0094] ​S301. For each of the multiple first sample objects, input the object portrait, the first sample tag set, and the second historical interest tag set of this first sample object into the initial model of this round of iteration, respectively obtain each second historical interest tag of each first sample object and the vector representation of the object portrait of this first sample object in this round of iteration, and the semantic similarity between each first sample tag and each second historical interest tag in this round of iteration;

[0095] S302. For each first sample tag of each first sample object, perform a weighted sum on the vector representations of each second historical interest tag in this round of iteration according to the semantic similarity between this first sample tag and each second historical interest tag in this round of iteration, to obtain the third vector of this first sample tag in this round of iteration;

[0096] S303. For each first sample tag of each first sample object, combine the third vector of this first sample tag in this round of iteration with the vector representation of the object portrait of the corresponding first sample object in this round of iteration, to obtain the fourth vector of this first sample tag in this round of iteration, and the fourth vector is used to characterize the interest degree of this first sample object in this first sample tag;

[0097] S304. For each first sample tag of each first sample object, obtain the recognition result of this first sample tag in this round of iteration output by the initial model of this round of iteration according to the fourth vector of this first sample tag in this round of iteration;

[0098] S305. Based on the difference degree between the recognition result of each first sample tag of each first sample object in this round of iteration and the exposure behavior of this first sample tag, adjust the model parameters of the initial model of this round of iteration to obtain the initial model of the next round of iteration.

[0099] In the embodiments of the present application, it is necessary to input the object portrait, the first sample tag set, and the second historical interest tag set of each first sample object into the initial model of the current iteration to obtain the vector representations of the object portrait and the second historical interest tags of the first sample object. For each first sample tag of the first sample object, determine the semantic similarity between the first sample tag and each second historical interest tag of the first sample object, use the semantic similarity as the weight of the corresponding second historical interest tag, and perform weighted summation on the vector representations of each second historical interest tag according to the weight to obtain the third vector of the first sample tag. Combine the third vector of the first sample tag with the vector representation of the object portrait of the first sample object corresponding to the first sample tag to obtain the fourth vector of the first sample tag. According to the fourth vector of the first sample tag, obtain the recognition result of each first sample tag in the first sample tag set, and then adjust the parameters of the initial model of the current iteration according to the exposure behavior of the first sample tag, and use the model with adjusted parameters as the initial model of the next iteration.

[0100] In the embodiments of the present application, the exposure behavior of the first sample tag includes a click behavior and a non-click behavior. If the exposure behavior of the user to the first sample tag is a click behavior, it indicates that the first sample tag is an interest tag of the user. On the contrary, it indicates that the first sample tag is not an interest tag of the user.

[0101] In some embodiments, the iteration stop condition may be:

[0102] The sum of the differences between the recognition results of each first sample tag of each first sample object and the exposure behavior of the corresponding first sample tag is less than a preset threshold; or

[0103] The number of iterations reaches a preset number.

[0104] Among them, the difference between the recognition result of a first sample tag of a first sample object and the exposure behavior of the first sample tag can be obtained according to the fourth vector of the first sample tag: input the fourth vector of the first sample tag into the sigmoid function to obtain the probability value that the first sample tag is an interest tag of the corresponding first sample object. If the exposure behavior of the first sample tag is a click behavior, then use 1 minus the probability value as the difference between the recognition result of the first sample tag and the exposure behavior of the first sample tag; if the exposure behavior of the first sample tag is a non-click behavior, then use the probability value as the difference between the recognition result of the first sample tag and the exposure behavior of the first sample tag.

[0105] Please refer to Figure 4 , which exemplarily shows the schematic flow chart of each round of iterative training provided by the embodiments of the present application.

[0106] In the embodiments of the present application, the initial model for each round of iteration includes a vector representation module, an identification module, and an adjustment model. The object portrait, the first sample label set, and the second historical interest label set of each first sample object among multiple first sample objects are input into the vector representation module to obtain the vector representation of the object portrait of each first sample object and the vector representation of the first historical interest label. Then, the vector representation of the object portrait of each first sample object, the vector representation of the first historical interest label, and the semantic similarity between each first sample label and each second historical interest label in each first sample object are input into the identification module to obtain the identification result of each first sample label of each first sample object, that is, the identification result set of the first sample label set of each first sample object is obtained. The identification result set of the first sample label set of each first sample object and the exposure behavior set of the corresponding first sample label set are input into the adjustment module, and the initial model parameters for this round of iteration are adjusted according to the difference degree between the identification result of the first sample label of each first sample object and the corresponding exposure behavior, so as to obtain the initial model for the next round of iteration.

[0107] Based on the above embodiments, as an optional embodiment, the type of the first sample label further includes a counter-intuitive interest label related to the portrait attribute pair:

[0108] The counter-intuitive interest label refers to: for the sample object group including the corresponding same portrait attribute pair, the label with a relatively small probability that the exposure behavior is a click behavior.

[0109] In the embodiments of the present application, the type of the first sample label of each first sample object is a common sense interest label and a counter-intuitive interest label. Each first sample label is defined as being related to the object attribute pair for both the common sense interest label and the counter-intuitive interest label. That is, for the sample object groups with different object attribute pairs, the common sense interest label and the counter-intuitive interest label may be different. For example, the common sense interest labels for the sample object group of "30-year-old + female" may be lipstick and dress, and the non-common sense interest label is football, while the common sense interest labels for the sample object group of "50-year-old + female" may be tissue and thermos cup, and the counter-intuitive interest label may be razor.

[0110] The counter-intuitive interest label related to an object attribute pair refers to: among the sample group including this portrait attribute pair, the label with a relatively small probability of being clicked or browsed after exposure. For example, the click behaviors of different labels for the sample group including the object attribute pair (age = 30, gender = male) are statistically analyzed to obtain one hundred labels. According to the click-through rate of each label, each label is sorted from large to small, and the last 20 labels are used as the counter-intuitive interest labels.

[0111] In the embodiments of the present application, the types of the first sample tags include common-sense interest tags and counter-intuitive interest tags. When training the initial model, training samples can be first based on the object portraits of multiple first sample objects, the counter-intuitive interest tags in the first sample tag set, and the second historical interest tag set, and training can be performed using the exposure behaviors of each counter-intuitive interest tag in the first sample tag set of the corresponding first sample object as training labels; then, training samples can be based on the object portraits of multiple first sample objects, the common-sense interest tags in the first sample tag set, and the second historical interest tag set, and the exposure behaviors of each common-sense interest tag in the first sample tag set of the corresponding first sample object can be used for the second training.

[0112] In the embodiments of the present application, considering the personalized interests and hobbies of each object, therefore, according to the counter-intuitive interest tags of the first sample objects and the corresponding exposure behaviors as training data, reinforcement learning is performed on the counter-intuitive interest tags of each first sample object, so that the trained interest tag recognition model can identify the personalized interest tags of the target object.

[0113] Based on the above embodiments, as an optional embodiment, the common-sense interest tags and counter-intuitive interest tags related to the portrait attribute pairs in the embodiments of the present application are determined by the following method:

[0114] S501. Obtain the object portraits of multiple second sample objects and the interest tag set of each second object in the first data source;

[0115] S502. For any portrait attribute pair, determine the sample object group including the portrait attribute pair from multiple second sample objects, and count the click-through rates of each interest tag in the interest tag set by the sample object group;

[0116] S503. For any portrait attribute pair, use the preset number of interest tags with the top click-through rates as the common-sense interest tags related to the portrait attribute pair, and use the interest tags in the interest tag set other than the common-sense interest tags as the counter-intuitive interest tags related to the portrait attribute pair.

[0117] Please refer to Figure 5 , which exemplarily shows the schematic flowchart of determining the common-sense interest tags and counter-intuitive interest tags provided by the embodiments of the present application.

[0118] In the embodiments of the present application, the first data source can be at least one of advertisement data, reading data, and ADX data. According to the behavior information of each second sample object in the first data source, the interest tag set of each second sample object is obtained, and at the same time, the object portrait of each second sample object is determined. As Figure 5As shown, object portraits and interest tag sets of the second sample object A, the second sample object B, and the second sample object C are obtained in the first data source.

[0119] In the embodiment of the present application, based on the object portraits of each of the above-determined second sample objects, a plurality of portrait attribute pairs are determined. A portrait attribute pair is two mutually related attributes in the object portrait. For example, (age = 30, gender = male) is a portrait attribute pair. For each portrait attribute pair, the sample object group under this portrait attribute pair is counted. It can be understood that the sample objects in this sample object group are the above-mentioned second sample objects. For each interest tag in the interest tag set of each second sample object in this sample object group, the click-through rate of all sample objects for this interest tag is counted as the click-through rate of this interest tag under this portrait attribute pair. As Figure 5 shown, taking the portrait attribute pair (age = 30, gender = male) as an example, after statistics, the sample object group under this portrait attribute pair includes the second sample object A and the second sample object C. That is to say, both the second sample object A and the second sample object C meet the conditions of age 30 and male gender. According to the union of the respective interest tag sets of the second sample object A and the second sample object C, interest tag 1, interest tag 2, interest tag 3, interest tag 4, and interest tag 5 are obtained. For each interest tag, the sum of the click-through rates of the second sample object A and the second sample object C for this interest tag is counted as the click-through rate of this interest tag under this portrait attribute pair.

[0120] In the embodiment of the present application, for any portrait attribute pair, all the interest tags under this portrait attribute pair are sorted in descending order of the click-through rate. The interest tags with higher click-through rates are used as the common sense interest tags under this portrait attribute pair, and the other interest tags except the common sense interest tags are used as the non-common sense interest tags under this portrait attribute pair. In the embodiment of the present application, having a higher click-through rate may mean that the ranking after sorting satisfies the 80 / 20 principle, that is, the top 80% of the interest tags after sorting are common sense interest tags and the bottom 20% of the interest tags are non-common sense interest tags. As Figure 5 shown, the click-through rates of each interest tag under the portrait attribute pair (age = 30, gender = male) under this portrait attribute pair are sorted. As Figure 5 shown, the result of sorting according to the click-through rate is interest tag 1 > interest tag 2 > interest tag 3 > interest tag 4 > interest tag 5. Therefore, according to the 80 / 20 principle, interest tag 1, interest tag 2, interest tag 3, and interest tag 4 are defined as the common sense interest tags under this portrait attribute pair, and interest tag 5 is defined as the non-common sense interest tag under this portrait attribute pair.

[0121] In some embodiments, the common-sense interest tags and counter-intuitive interest tags may also be divided according to other rules, as long as the common-sense interest tags can reflect that the object attribute is of interest to most objects, and the counter-intuitive interest tags can reflect that the object attribute is of interest to a small number of objects.

[0122] In the embodiments of the present application, by obtaining the object portraits and the corresponding interest tag sets of multiple second sample objects in the first data source, the sample object groups corresponding to each portrait attribute in the object portrait are obtained. According to the interest tag sets of all the second sample objects in the sample object group corresponding to each portrait attribute, the correspondence between each portrait attribute pair and each interest tag is obtained. According to the click-through rate of each interest tag corresponding to each portrait attribute pair, the common-sense interest tags and counter-intuitive interest tags corresponding to each portrait attribute pair are determined. The common-sense interest tags are used to reflect the tags that most second sample objects in the sample group corresponding to the corresponding portrait attribute pair are interested in, and the counter-intuitive interest tags are used to reflect the tags that a small number of second sample objects in the sample group corresponding to the corresponding portrait attribute pair are interested in.

[0123] Based on the above embodiments, as an alternative embodiment, the first sample tag set of each first sample object in the embodiments of the present application is determined by the following method:

[0124] S601. Obtain the object portraits of multiple first sample objects, the multiple third sample tags of each first sample object, and the exposure behaviors of each third sample tag in the second data source;

[0125] S602. For each portrait attribute of each first sample object, determine the common-sense interest tags and counter-intuitive interest tags related to the portrait attribute. If the first sample object has a third sample tag that is the same as the common-sense interest tag or the counter-intuitive interest tag, update the third sample tag to the corresponding common-sense interest tag or counter-intuitive interest tag.

[0126] Please refer to Figure 6 , which exemplarily shows a schematic flow chart for determining the first sample tag set of each first sample object provided by the embodiments of the present application.

[0127] In the embodiments of the present application, the second data source is a data source different from the first data source. For example, if the first data source is advertising data, the second data source may be at least one of reading data and ADX data. According to the data in the second data source, the object portraits of multiple first sample objects and the third sample tag set of each first sample object are determined. The third sample tag set includes multiple third sample tags and the exposure behaviors of each third sample tag. As Figure 6As shown, for a first sample object, the object portrait of the first sample object is (age = 30, gender = male, job type = doctor), and the third sample labels in the third sample label set of the first sample object are (label 1, label 2, label 3, label 4, label 5, label 6, label 7, label 8, label 9). It should be noted that each third sample label has corresponding exposure behavior information.

[0128] In an embodiment of the present application, for one of the multiple first sample objects described above, according to each portrait attribute of the first sample object, as well as the common sense interest labels and anti-common sense interest labels corresponding to each portrait attribute determined in step S503, the common sense interest labels and anti-common sense interest labels related to the portrait attribute are determined. If the common sense interest label or anti-common sense interest label is a third sample label of the first sample object, then the third sample label is updated to the corresponding common sense interest label or anti-common sense interest label.

[0129] In an embodiment of the present application, for each portrait attribute of a first sample object, the method for determining the common sense interest labels and anti-common sense interest labels related to the portrait attribute can be to combine multiple portrait attributes of the object portrait of the first sample object in pairs to form multiple portrait attribute pairs. For each portrait attribute pair, by combining the common sense interest labels and anti-common sense interest labels corresponding to the portrait attribute pair obtained in step S503, the intersection of the third sample label of the first sample object and the common sense interest labels and anti-common sense interest labels corresponding to the portrait attribute pair is used as the common sense interest labels and anti-common sense interest labels of the first sample object corresponding to the portrait attribute pair. The common sense interest labels of the first sample object corresponding to all portrait attribute pairs are used as the common sense interest labels of the first sample object, and the anti-common sense interest labels of the first sample object corresponding to all portrait attributes are used as the anti-common sense interest labels of the first sample object. The common sense interest labels and anti-common sense interest labels of the first sample object together form the first sample label set of the first sample object. It should be noted that if a third sample label is both a common sense interest label of the first sample object and an anti-common sense interest label of the first sample object, then the third sample label is defined as an anti-common sense interest label.

[0130] As Figure 6 shown, Figure 6Intuitively shows the process of determining the first sample label set of the first sample object above. According to the pairwise combination of the mutual attributes of the first sample object, three portrait attribute pairs are obtained: (age = 30, gender = male), (age = 30, job type = doctor), and (gender = male, job type = doctor). For the common sense interest labels under the portrait attribute pair (age = 30, gender = male) in S403, they are label 1 and label 2, and the anti-common sense interest label is label 3. Thus, the common sense interest labels of the first sample object under (age = 30, gender = male) are label 1 and label 2, and the anti-common sense interest label is label 3; for the common sense interest labels under the portrait attribute pair (age = 30, job type = doctor), they are label 4, label 5, and label 10, and the anti-common sense interest labels are label 6 and label 11. Then, the common sense interest labels of the first sample object under (age = 30, job type = doctor) are label 4 and label 5, and the anti-common sense interest label is label 6; for the common sense interest labels under the portrait attribute pair (gender = male, job type = doctor), they are label 3 and label 7, and the anti-common sense interest label is label 6. Then, the common sense interest labels of the first sample label under (gender = male, job type = doctor) are label 3 and label 7, and the anti-common sense interest label is label 6. Take the union of the common sense interest labels under the three portrait attribute pairs of the first sample object to obtain (label 1, label 2, label 3, label 4, label 5, label 7), and take the union of the anti-common sense interest labels under the three mutual attribute pairs of the first sample object to obtain (label 3, label 6). Then, take label 1, label 2, label 4, label 5, and label 7 as the common sense interest labels of the first sample object, and label 3, label 6 as the anti-common sense interest labels of the first sample object. The common sense interest labels and anti-common sense interest labels of the first sample object constitute the first sample label set of the first sample object.

[0131] In the embodiment of the present application, for the first sample interest label of the first sample object used to train the initial model, the determination of the first sample interest label comprehensively considers the object portrait information and behavior information of the first sample object, as well as the object portrait information and behavior information of multiple second sample objects, so that the trained interest label recognition model can accurately identify whether the label to be recognized is the interest label of the target object according to the object portrait of the target object.

[0132] Based on the above embodiments, as an optional embodiment, the label to be recognized, the first historical interest label, the first sample label, and the second historical interest label include at least two levels of label attributes.

[0133] In the embodiments of the present application, the format of each tag is the same, and each includes at least two levels of tag attributes, with each level being the category information of the commodity. For example, when the tag includes two levels of tag attributes, a certain brand of smartphone can be represented as <Electronic Devices, Mobile Phones>; when the tag includes three levels of tag attributes, a certain brand of smartphone can be represented as <Electronic Devices, Mobile Phones, Smartphones>; when the tag includes four levels of tag attributes, a certain brand of smartphone can be represented as <Electronic Devices, Mobile Phones, Smartphones, Music Smartphones>.

[0134] It can be understood that if an object was previously interested in clothing, shoes, and bags, then when a tag of "spinning skirt" appears, the object is likely to be interested. If an object was not previously interested in alcoholic products, then when a baijiu tag is suddenly added, the object is likely not to be interested. Therefore, the tag format in the embodiments of the present application includes at least two levels of tag attributes, which can obtain a more accurate semantic similarity between tags compared to when the tag only includes one tag attribute, making the determination result of the target object's interest tags more accurate.

[0135] In some optional embodiments, each tag also includes product attributes and brand attributes. For example, a certain brand of smartphone is represented as <Electronic Devices, Mobile Phones, Smartphones, Music Smartphones, Mobile Phone Product Name, Brand Name>.

[0136] In other optional embodiments, for multiple first historical interest tags in the first historical interest tag set, the multiple first historical interest tags can be combined according to the tag attributes of the highest level, thereby reducing the number of tags when inputting into the interest tag model. For example, for the following four first historical interest tags:

[0137] <Clothing, Shoes & Bags, Men's Fashion Shoes, Slippers, Flip Flops, Summer Flip Flops, Brand 1>

[0138] <Clothing, Shoes & Bags, Women's Fashion Shoes, Casual Shoes, Flat Shoes, Autumn Flat Shoes, Brand 2>

[0139] <Digital Devices, Photography & Videography, DSLR Cameras, Medium Format DSLR Cameras, High - End Medium Format DSLR Cameras, Brand 3>

[0140] <Digital Devices, Mobile Phones & Telephones, Mobile Phones, Smartphones, Music Smartphones, Brand 4>

[0141] It can be combined into the following tags <Apparel, Shoes & Bags, Digital Devices; Men's Fashion Shoes, Women's Fashion Shoes, Photography & Videography, Mobile Phones & Telephones; Slippers, Casual Shoes, DSLR Cameras, Mobile Phones; Flip Flops, Flats, Medium Format DSLR Cameras, Smartphones; Summer Flip Flops, Autumn Flats, High - end Medium Format DSLR Cameras, Music Smartphones; Brand 1, Brand 2, Brand 3, Brand 4>. It should be noted that when inputting in this tag format into the interest tag recognition model, the vector representations of the above four first - historical interest tags are obtained.

[0142] Similarly, the format of the second - historical interest tags can also be set to the above - mentioned format, which will not be elaborated here.

[0143] Please refer to Figure 7 , which exemplarily shows a flowchart of the method for determining the interest tags of the object in another embodiment of the present application, as Figure 7 shown, the embodiments of the present application are mainly divided into two parts: the first part is the training of the interest tag recognition model, and the second part is to recognize the tags to be recognized of the target object according to the interest tag recognition model to obtain the recognition result of the tags to be recognized of the target object. Specifically:

[0144] S801. Obtain the object portraits of multiple second - sample objects and the interest tag sets of each second - sample object in the first data source;

[0145] S802. For any portrait - attribute pair, determine the sample - object group including this portrait - attribute pair from multiple second - sample objects, and count the click - through rates of each interest tag in the interest tag set of this sample - object group;

[0146] S803. For any portrait - attribute pair, take the preset number of interest tags with the top click - through rates as the common - sense interest tags related to this portrait - attribute pair, and take the interest tags in the interest tag set other than this common - sense interest tag as the anti - common - sense interest tags related to this portrait - attribute pair;

[0147] S804. Obtain the object portraits of multiple first - sample objects, multiple third - sample tags of each first - sample object, and the exposure behaviors of each third - sample tag in the second data source;

[0148] S805. For each portrait attribute of the first - sample object, determine the common - sense interest tags and anti - common - sense interest tags related to this portrait attribute. If the first - sample object has a third - sample tag that is the same as this common - sense interest tag or anti - common - sense interest tag, update this third - sample tag to the corresponding common - sense interest tag or anti - common - sense interest tag;

[0149] S806. Using the object portraits of the multiple first sample objects, the first sample label set, and the second historical interest label set as training samples, and using the exposure behaviors of each first sample label in the first sample label set of the corresponding first sample object as training labels, perform multiple rounds of iterative training on the initial model until the iteration stop condition is met; wherein, the first sample label set includes common sense interest labels and counter - common sense interest labels;

[0150] S807. Determine the object portrait, the label set to be recognized, and the first historical interest label set of the target object;

[0151] S808. Input the object portrait, the label set to be recognized, and the first historical interest label set of the target object into the pre - trained interest label recognition model, and obtain the recognition result set output by the interest label recognition model. The recognition result set includes the recognition results of whether each label to be recognized is an interest label of the target object.

[0152] Please refer to Figure 8 , which exemplarily shows a schematic diagram of the application scenario of a method for determining the interest labels of an object in an embodiment of the present application. This scenario is the advertisement recommendation scenario. As Figure 8 shown, the scenario includes object 801, the terminal 802 used by object 801, the server 803 storing the interest label recognition model, the server 804 storing multiple historical advertisements (i.e., the first historical interest labels) that object 801 is interested in, the server 805 storing the object portrait of object 801, and the server 806 storing multiple advertisements (i.e., the labels to be recognized) to be recommended to object 801.

[0153] The terminal 802 includes a page for displaying advertisements. When object 801 enters this page, the terminal 802 applies to the server 803 to push advertisement information. After receiving the application from the terminal 802, the server 803 sends a request to obtain multiple historical advertisements that object 801 is interested in to the server 804, sends a request to obtain the object portrait of object 801 to the server 805, and sends multiple advertisements to be recommended to object 801 to the server 806. In response to the requests from the server 803, the server 804, the server 805, and the server 806 respectively send multiple historical advertisements that object 801 is interested in, the object portrait of object 801, and multiple advertisements to be recommended to object 801 to the server 803. The server 803 uses the interest label recognition model to recognize multiple advertisements to be pushed to object 801, determines the target advertisements that object 801 is interested in, and pushes them to the terminal 802 used by object 801. The terminal 802 receives the target advertisements sent by the server 803 and displays them on the above - mentioned page.

[0154] The embodiment of the present application provides a device for determining the interest labels of an object, as Figure 9As shown in the figure, the apparatus for determining the interest tags of the object may include: an obtaining module 901 and an input module 902. Among them,

[0155] The obtaining module 901 is configured to determine the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object;

[0156] The input module 902 is configured to input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model, and obtain a recognition result set output by the interest tag recognition model. The recognition result set includes the recognition results of whether each tag to be recognized is an interest tag of the target object;

[0157] Among them, the interest tag recognition model is trained with the object portraits of multiple first sample objects, the first sample tag set, and the second historical interest tag set as training samples, and the exposure behaviors of each first sample tag in the first sample tag set of the corresponding first sample object as training labels;

[0158] Among them, the types of the first sample tags include common sense interest tags and counter-intuitive interest tags related to portrait attribute pairs;

[0159] The common sense interest tag refers to a tag with a relatively high probability of having a click behavior as the exposure behavior for a group of sample objects including the corresponding same portrait attribute pair;

[0160] The counter-intuitive interest tag refers to a tag with a relatively low probability of having a click behavior as the exposure behavior for a group of sample objects including the corresponding same portrait attribute pair.

[0161] The apparatus for determining the interest tags of the object in the embodiment of the present application can execute the method for determining the interest tags of the object provided in the embodiment of the present application, and its implementation principle is similar. The actions performed by each module in the apparatus for determining the interest tags of the object in each embodiment of the present application correspond to the steps in the method for determining the interest tags of the object in each embodiment of the present application. For the detailed function descriptions of each module of the apparatus for determining the interest tags of the object, reference can be specifically made to the descriptions in the corresponding methods shown above, and details are not described herein again.

[0162] In an alternative embodiment, an electronic device is provided, as Figure 10 shown, Figure 10The electronic device 4000 shown includes: a processor 4001 and a memory 4003. Among them, the processor 4001 and the memory 4003 are connected, such as being connected through a bus 4002. Optionally, the electronic device 4000 may further include a transceiver 4004, and the transceiver 4004 can be used for data interaction between this electronic device and other electronic devices, such as data sending and / or data receiving, etc. It should be noted that in practical applications, the transceiver 4004 is not limited to one, and the structure of the electronic device 4000 does not constitute a limitation to the embodiments of the present application.

[0163] The processor 4001 can be a CPU (Central Processing Unit, central processor), a general-purpose processor, a DSP (Digital Signal Processor, data signal processor), an ASIC (Application Specific Integrated Circuit, application-specific integrated circuit), an FPGA (Field Programmable Gate Array, field programmable gate array), or other programmable logic devices, transistor logic devices, hardware components, or any combination thereof. It can implement or execute various exemplary logic blocks, modules, and circuits described in combination with the disclosure of the present application. The processor 4001 can also be a combination that realizes computing functions, such as a combination including one or more microprocessors, a combination of a DSP and a microprocessor, etc.

[0164] The bus 4002 may include a path for transmitting information between the above components. The bus 4002 can be a PCI (Peripheral Component Interconnect, peripheral component interconnect standard) bus or an EISA (Extended Industry Standard Architecture, extended industry standard structure) bus, etc. The bus 4002 can be divided into an address bus, a data bus, a control bus, etc. For the sake of representation, Figure 10 only a thick line is used to represent it in the figure, but it does not mean that there is only one bus or one type of bus.

[0165] The memory 4003 can be a ROM (Read Only Memory), or other types of static storage devices that can store static information and instructions, a RAM (Random Access Memory), or other types of dynamic storage devices that can store information and instructions. It can also be an EEPROM (Electrically Erasable Programmable Read Only Memory), a CD-ROM (Compact Disc Read Only Memory), or other optical disc storage, optical disc storage (including compact discs, laser discs, optical discs, digital versatile discs, Blu-ray discs, etc.), magnetic disk storage media, other magnetic storage devices, or any other medium that can be used to carry or store computer programs and can be read by a computer, which is not limited herein.

[0166] The memory 4003 is used to store the computer program for implementing the embodiments of the present application and is controlled by the processor 4001 for execution. The processor 4001 is used to execute the computer program stored in the memory 4003 to implement the steps shown in the foregoing method embodiments.

[0167] Among them, the electronic device package can include, but is not limited to, mobile terminals such as mobile phones, laptop computers, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Tablet Computers), PMPs (Portable Multimedia Players), in-vehicle terminals (such as in-vehicle navigation terminals), etc., and fixed terminals such as digital TVs, desktop computers, etc. Figure 10 The electronic device shown is only an example and should not impose any limitations on the functions and scope of use of the embodiments of the present disclosure.

[0168] An embodiment of the present application provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps and corresponding contents of the foregoing method embodiment can be implemented. Compared with the prior art, it can be achieved that: by inputting the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model, an identification result set output by the interest tag recognition model is obtained. Since the interest tag recognition model uses the object portraits, the first sample tag set, and the second historical interest tag set of multiple first sample objects as training samples, and uses the exposure behavior of each first sample tag in the first sample tag set of the corresponding sample object as a training label, the types of the first sample tags include common sense interest tags, and the common sense interest tags are related to portrait attribute pairs, that is, the first sample tags of different portrait attribute pairs are different. The present application statistically analyzes the click behavior of the sample object group at the granularity of portrait attribute pairs to obtain common sense interest tags. The model can analyze the interest tags in each common sense interest tag of the sample object by analyzing the object portrait, the common sense interest tags, the corresponding exposure behavior, and the second historical interest tag set of the sample object. Furthermore, after the model training is completed, it is possible to accurately determine, based on the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object, the recognition result of whether the tag to be recognized is an interest tag of the target object, that is, it is possible to determine the recognition result of whether each tag to be recognized generated by multiple data sources is an interest tag of the target object.

[0169] It should be noted that the above-mentioned computer-readable medium in the present disclosure can be a computer-readable signal medium, a computer-readable storage medium, or any combination of the two. The computer-readable storage medium can be, for example, but not limited to, an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the above. More specific examples of the computer-readable storage medium can include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the above. In the present disclosure, the computer-readable storage medium can be any tangible medium that contains or stores a program, which can be used by or in conjunction with an instruction execution system, apparatus, or device. And in the present disclosure, the computer-readable signal medium can include a data signal propagated in a baseband or as part of a carrier wave, which carries computer-readable program code. Such a propagated data signal can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. The computer-readable signal medium can also be any computer-readable medium other than the computer-readable storage medium, which can send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any appropriate medium, including but not limited to: wires, optical cables, RF (radio frequency), etc., or any suitable combination of the above.

[0170] The embodiments of the present application also provide a computer program product, including a computer program, which when executed by a processor can implement the steps and corresponding content of the foregoing method embodiments. Compared with the prior art, it can be achieved that: by inputting the object portrait of the target object, the set of tags to be recognized, and the first historical interest tag set into a pre-trained interest tag recognition model, an identification result set output by the interest tag recognition model is obtained. Since the interest tag recognition model uses the object portraits of multiple first sample objects, the first sample tag set, and the second historical interest tag set as training samples, and uses the exposure behavior of each first sample tag in the first sample tag set of the corresponding sample object as the training label, the types of the first sample tags include common sense interest tags, and the common sense interest tags are related to the portrait attribute pairs. That is, the first sample tags of different portrait attribute pairs are different. The present application statistically analyzes the click behavior of the sample object group at the granularity of the portrait attribute pairs to obtain common sense interest tags. The model can analyze the interest tags in each common sense interest tag of the sample object by analyzing the object portrait of the sample object, the common sense interest tags, the corresponding exposure behavior, and the second historical interest tag set. Then, after the model training is completed, it can accurately determine, based on the object portrait of the target object, the set of tags to be recognized, and the first historical interest tag set, the recognition result of whether the tag to be recognized is an interest tag of the target object, that is, it can determine the recognition result of whether each tag to be recognized generated by multiple data sources is an interest tag of the target object.

[0171] The terms "first", "second", "third", "fourth", "1", "2", etc. (if any) in the description and claims of the present application and the above drawings are used to distinguish similar objects and do not necessarily need to be used to describe a specific order or sequence. It should be understood that such data can be interchanged under appropriate circumstances so that the embodiments of the present application described herein can be implemented in an order other than that shown or described in words.

[0172] It should be understood that although the flowcharts of the embodiments of the present application indicate various operation steps by arrows, the execution order of these steps is not limited to the order indicated by the arrows. Unless there is a clear description in this article, in some implementation scenarios of the embodiments of the present application, the implementation steps in each flowchart can be executed in other orders according to requirements. In addition, some or all of the steps in each flowchart may include multiple sub-steps or multiple stages based on the actual implementation scenario. Some or all of these sub-steps or stages can be executed at the same time, and each sub-step or stage in these sub-steps or stages can also be executed at different times. In the scenario where the execution times are different, the execution order of these sub-steps or stages can be flexibly configured according to requirements, and the embodiments of the present application do not limit this.

[0173] The above are only optional implementation manners of some implementation scenarios of the present application. It should be noted that for those of ordinary skill in the art, without departing from the technical concept of the solution of the present application, other similar implementation means based on the technical idea of the present application also fall within the protection scope of the embodiments of the present application.

Claims

1. A method for determining interest tags of an object, characterized in that Including: Determine the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object; Input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain a recognition result set output by the interest tag recognition model, where the recognition result set includes the recognition results of whether each tag to be recognized is an interest tag of the target object; Among them, the interest tag recognition model is trained with the object portraits, the first sample tag sets, and the second historical interest tag sets of multiple first sample objects as training samples, and the exposure behaviors of each first sample tag in the first sample tag sets of the corresponding first sample objects as training labels; Among them, the types of the first sample tags include common sense interest tags related to portrait attribute pairs; The common sense interest tag refers to: for a group of sample objects including the same portrait attribute pair, a tag with a relatively high probability that the exposure behavior is a click behavior.

2. The method according to claim 1, wherein The step of inputting the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain a recognition result set output by the interest tag recognition model includes: Input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model to obtain vector representations of each first historical interest tag and the object portrait of the target object, and the semantic similarity between each tag to be recognized and each first historical interest tag; For each tag to be recognized, perform weighted summation on the vector representations of each first historical interest tag according to the semantic similarity between the tag to be recognized and each first historical interest tag to obtain a first vector of the tag to be recognized; For each tag to be recognized, combine the first vector of the tag to be recognized and the vector representation of the object portrait of the target object to obtain a second vector of the tag to be recognized, and the second vector is used to characterize the interest degree of the target object in the tag to be recognized; For each tag to be recognized, obtain the recognition result of the tag to be recognized output by the interest tag recognition model according to the second vector of the tag to be recognized.

3. The method according to claim 1 or 2, characterized in that, The training method of the interest tag recognition model includes: Use the object portraits, the first sample tag sets, and the second historical interest tag sets of the multiple first sample objects as training samples, and use the exposure behaviors of each first sample tag in the first sample tag sets of the corresponding first sample objects as training labels to perform multiple rounds of iterative training on the initial model until the iterative stop condition is met; Among them, each round of iterative training includes: For each first sample object, input the object portrait, the first sample tag set, and the second historical interest tag set of the first sample object into the initial model of this round of iteration, and respectively obtain each second historical interest tag of the first sample object and the vector representation of the object portrait of the first sample object in this round of iteration, and the semantic similarity between each first sample tag and each second historical interest tag in this round of iteration; For each first sample label of each first sample object, perform a weighted sum on the vector representations of each second historical interest label in this iteration according to the semantic similarity between the first sample label and each second historical interest label in this iteration, to obtain the third vector of the first sample label in this iteration; For each first sample label of each first sample object, combine the third vector of the first sample label in this iteration with the vector representation of the object portrait of the first sample object in this iteration, to obtain the fourth vector of the first sample label in this iteration, where the fourth vector is used to characterize the degree of interest of the first sample object in the first sample label; For each first sample label of each first sample object, obtain the recognition result of the first sample label in this iteration output by the initial model of this iteration according to the fourth vector of the first sample label in this iteration; Based on the difference degree between the recognition result of each first sample label of each first sample object in this iteration and the exposure behavior of the first sample label, adjust the model parameters of the initial model of this iteration to obtain the initial model of the next iteration.

4. The method according to claim 3, wherein The type of the first sample label further includes a counter-intuitive interest label related to a portrait attribute pair: The counter-intuitive interest label refers to: for a sample object group including the corresponding same portrait attribute pair, a label with a relatively low probability that the exposure behavior is a click behavior.

5. The method according to any one of claims 1 to 3, characterized in that The common-sense interest label and the counter-intuitive interest label related to the portrait attribute pair are determined by the following method: Obtain the object portraits of multiple second sample objects and the interest label set of each second sample object in a first data source; For any portrait attribute pair, determine the sample object group including the portrait attribute pair from the multiple second sample objects, and count the click-through rate of each interest label in the interest label set by the sample object group; For any portrait attribute pair, use the preset number of interest labels with the top click-through rates as the common-sense interest labels related to the portrait attribute pair, and use the interest labels in the interest label set other than the common-sense interest labels as the counter-intuitive interest labels related to the portrait attribute pair.

6. The method according to claim 4, wherein The first sample label set of each first sample object is determined by the following method: Obtain the object portraits of multiple first sample objects, multiple third sample labels of each first sample object, and the exposure behavior of each third sample label in a second data source; For each portrait attribute of the first sample object, determine the common-sense interest label and the counter-intuitive interest label related to the portrait attribute. If the first sample object has a third sample label that is the same as the common-sense interest label or the counter-intuitive interest label, update the third sample label to the corresponding common-sense interest label or counter-intuitive interest label.

7. According to the method described in any one of claims 1-6, any one of the to-be-recognized label, the first sample label, the second sample label, the third sample label, the first historical interest label, and the second historical interest label includes at least two levels of label attributes.

8. An apparatus for determining interest tags of an object, characterized in that, Including: An obtaining module, configured to determine an object portrait, a set of tags to be recognized, and a first historical interest tag set of a target object; An input module, configured to input the object portrait, the set of tags to be recognized, and the first historical interest tag set of the target object into a pre-trained interest tag recognition model, and obtain a recognition result set output by the interest tag recognition model, where the recognition result set includes recognition results of whether each tag to be recognized is an interest tag of the target object; wherein the interest tag recognition model is trained with the object portraits of multiple first sample objects, a first sample tag set, and a second historical interest tag set as training samples, and the exposure behaviors of each first sample tag in the first sample tag set of the corresponding first sample object as training labels; wherein the types of the first sample tags include common sense interest tags and counter-intuitive interest tags related to portrait attribute pairs; The common sense interest tag refers to a tag with a relatively high probability that the exposure behavior is a click behavior for a sample object group including the corresponding same portrait attribute pair; The counter-intuitive interest tag refers to a tag with a relatively low probability that the exposure behavior is a click behavior for a sample object group including the corresponding same portrait attribute pair.

9. An electronic device, comprising a memory, a processor, and a computer program stored on the memory, characterized in that, The processor executes the computer program to implement the steps of the method according to any one of claims 1-7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1-7.

11. A computer program product, comprising a computer program, characterized in that, When the computer program is executed by the processor, it implements the steps of the method according to any one of claims 1-7.

Images