Account portrait information generation method and device, equipment, storage medium and product

Abstract

The invention discloses an account portrait information generation method and device, equipment, a storage medium and a product, and belongs to the technical field of artificial intelligence. According to the account portrait information generation method, the information under the time dimension, the behavior dimension and the data source dimension in the account behavior record can be effectively fused to obtain the account portrait information. And in the fusion process, objective and universal fusion weights are set for the information under the three dimensions, so that the quality of the generated account portrait information is improved. Moreover, by improving the setting mode of the weight, the dependence on manpower is reduced, so that the generated account portrait information has universality, that is to say, good effects can be achieved when related services are provided based on the account portrait information in different business scenes. In addition, the account portrait information generation method also considers the information under the data source dimension, so that the quality of the account portrait information is further improved.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence technology, and in particular to a method, device, equipment, storage medium and product for generating account portrait information. Background Art

[0002] Applications in related technologies can predict the interest or behavior tendencies of the user corresponding to the account through account portrait information, thereby providing corresponding services to the user corresponding to the account. The ability of account portrait information to represent the interest or behavior tendencies of the user corresponding to the account is closely related to the service content provided by the application, and also affects the service quality of the service that the application can provide. Therefore, many applications in related technologies are committed to optimizing account portrait information and improving the quality of account portrait information. However, at present, the quality of account portrait information generated by related technologies still needs to be improved. Summary of the invention

[0003] The embodiments of the present application provide a method, apparatus, device, storage medium and product for generating account portrait information, which can significantly improve the quality of account portrait information, that is, improve the ability of account portrait information to represent the interest tendencies or behavioral tendencies of the user corresponding to the account.

[0004] According to one aspect of an embodiment of the present application, a method for generating account portrait information is provided, the method comprising:

[0005] Acquire account behavior records related to the account, wherein the account behavior records include a time dimension tag, a behavior dimension tag, and a data source dimension tag, wherein the behavior dimension tag indicates a category associated with a behavior performed by the corresponding account, and the data source dimension tag indicates a source of the behavior performed by the corresponding account;

[0006] Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account portrait information is generated;

[0007] Determining a second weight corresponding to the behavior dimension label according to a significance index corresponding to the category pointed to by the behavior dimension label;

[0008] Determining, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag;

[0009] The account portrait information is generated according to the account behavior record, the first weight, the second weight and the third weight.

[0010] According to one aspect of an embodiment of the present application, a device for generating account portrait information is provided, the device comprising:

[0011] An account data extraction module, used to obtain account behavior records related to an account, wherein the account behavior records include a time dimension tag, a behavior dimension tag, and a data source dimension tag, wherein the behavior dimension tag indicates a category associated with a behavior performed by a corresponding account, and the data source dimension tag indicates a source of generation of the behavior performed by the corresponding account;

[0012] The account profile information generation module is used to perform the following operations:

[0013] Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account portrait information is generated;

[0014] Determining a second weight corresponding to the behavior dimension label according to a significance index corresponding to the category pointed to by the behavior dimension label;

[0015] Determining, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag;

[0016] The account portrait information is generated according to the account behavior record, the first weight, the second weight and the third weight.

[0017] According to one aspect of an embodiment of the present application, a computer device is provided, which includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by the processor to implement the above-mentioned account portrait information generation method.

[0018] According to one aspect of an embodiment of the present application, a computer-readable storage medium is provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. The at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by a processor to implement the above-mentioned account portrait information generation method.

[0019] According to one aspect of the embodiments of the present application, a computer program product is provided, the computer program product comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes to implement the above-mentioned account portrait information generation method.

[0020] The technical solution provided in the embodiments of the present application can bring the following beneficial effects:

[0021] The embodiment of the present application provides a method for generating account portrait information, which can effectively fuse the information in the time dimension, behavior dimension and data source dimension in the account behavior record to obtain account portrait information. In the process of fusion, objective and universal fusion weights are set for the information in these three dimensions, thereby improving the quality of the generated account portrait information. In addition, by improving the setting method of the weight, the dependence on manual work is reduced, so that the generated account portrait information has universality, that is, providing related services based on the account portrait information in different business scenarios can achieve good results.

[0022] The account portrait information generation method also takes into account the information under the data source dimension, thereby incorporating the data source factor into the considerations when generating account portrait information, further improving the quality of the account portrait information. BRIEF DESCRIPTION OF THE DRAWINGS

[0023] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0024] Figure 1 It is a schematic diagram of an application program running environment related to a method provided by an embodiment of the present application;

[0025] Figure 2 It is a flowchart of a method for generating account portrait information provided by an embodiment of the present application;

[0026] Figure 3 This is a schematic diagram of an account behavior record space provided by an embodiment of the present application;

[0027] Figure 4 This is a schematic diagram of an account-category portrait provided by an embodiment of the present application;

[0028] Figure 5 It is a schematic diagram of another method for generating account portrait information provided by an embodiment of the present application;

[0029] Figure 6 This is a schematic diagram of an information expansion method in flat fusion provided by an embodiment of the present application;

[0030] Figure 7 It is a schematic diagram of a feature expansion effect provided by an embodiment of the present application;

[0031] Figure 8 This is a schematic diagram of timing requirements for constructing behavior samples provided by an embodiment of the present application;

[0032] Fig. 9 It is a block diagram of an account portrait information generating device provided by an embodiment of the present application;

[0033] Fig.10 It is a structural block diagram of a computer device provided by an embodiment of the present application. DETAILED DESCRIPTION

[0034] Before introducing the method embodiments provided in the present application, a brief introduction is first given to the relevant terms or nouns that may be involved in the method embodiments of the present application to facilitate understanding by technicians in the field of the present application.

[0035] Artificial Intelligence (AI) is the theory, method, technology and application system that uses digital computers or machines controlled by digital computers to simulate, extend and expand human intelligence, perceive the environment, acquire knowledge and use knowledge to obtain the best results. In other words, artificial intelligence is a comprehensive technology in computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can respond in a similar way to human intelligence. Artificial intelligence is to study the design principles and implementation methods of various intelligent machines so that machines have the functions of perception, reasoning and decision-making.

[0036] Artificial intelligence technology is a comprehensive discipline that covers a wide range of fields, including both hardware-level and software-level technologies. The basic technologies of artificial intelligence generally include sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technology, operation / interaction systems, mechatronics and other technologies. Artificial intelligence software technology mainly includes computer vision technology, speech processing technology, natural language processing technology, and machine learning / deep learning.

[0037] Machine Learning (ML) is a multi-disciplinary subject that involves probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and other disciplines. It specializes in studying how computers simulate or implement human learning behavior to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning is the core of artificial intelligence and the fundamental way to make computers intelligent. Its applications are spread across all areas of artificial intelligence. Machine learning and deep learning usually include artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and self-learning.

[0038] Deep learning: The concept of deep learning originates from the study of artificial neural networks. A multi-layer perceptron with multiple hidden layers is a deep learning structure. Deep learning discovers distributed feature representations of data by combining low-level features to form more abstract high-level representations of attribute categories or features.

[0039] Cloud technology refers to a hosting technology that unifies hardware, software, network and other resources in a wide area network or local area network to achieve data computing, storage, processing and sharing. Cloud technology is a general term for network technology, information technology, integration technology, management platform technology, application technology, etc. based on the cloud computing business model. It can form a resource pool, which is used on demand and is flexible and convenient. Cloud computing technology will become an important support. The background services of technical network systems require a large amount of computing and storage resources, such as video websites, picture websites and more portal websites. With the high development and application of the Internet industry, in the future, each item may have its own identification mark, and all need to be transmitted to the background system for logical processing. Data of different levels will be processed separately. All kinds of industry data require strong system backing support, which can only be achieved through cloud computing.

[0040] BERT (Bidirectional Encoder Representation from Transformers) is a large-scale text pre-training model. BERT uses a 12-layer transformer encoder to significantly improve the benchmark performance of natural language processing tasks. Compared with word2vec (word vector), BERT, which has been pre-trained on massive amounts of text, can introduce more transfer knowledge into the classification algorithm and provide more accurate text features.

[0041] Transformer: is a neural network that learns context and therefore meaning by extracting relationships in sequence data. Transformer models apply a set of evolving mathematical techniques called attention or self-attention to detect subtle ways in which even distant data elements in a series influence and depend on each other.

[0042] TGI: TGI (Target Group Index), or target group index, is an index that reflects the strength or weakness of the target group within a specific research scope. TGI index = [the proportion of groups with a certain characteristic in the target group / the proportion of groups with the same characteristic in the overall population] * standard number 100.

[0043] User portraits, also known as user roles, are an effective tool for outlining target users and linking user demands with design directions. User portraits have been widely used in various fields. In the actual operation process, the user's attributes, behaviors, and expected data conversions are often linked. As a virtual representative of the actual user, the user role formed by the user's portrait needs to be representative and can represent the main audience and target group of the object to which it corresponds. In the embodiments of the present application, account portrait information is used to represent the user portrait of the user corresponding to the account.

[0044] Feature crossing: For AI algorithm models that use past and current information to predict the future, feature crossing essentially means that the features contain future information. For the online reasoning process, the information used to construct features can only come from the present or the past, so there is naturally no feature crossing problem. For the offline training process, when constructing features, information after the sample occurs may be mistakenly introduced, resulting in feature crossing.

[0045] Bias indicators (also known as bias indicators) are a type of indicator used to measure the accuracy of algorithms in the fields of machine learning, deep learning, and data mining. Bias indicators include accuracy, supplementary accuracy, precision, recall, and so on.

[0046] Before describing the embodiments of the present application in detail, the relevant technical background related to the embodiments of the present application is introduced to facilitate understanding by those skilled in the art in the field of the present application.

[0047] The application in the related technology can predict the interest tendency or behavior tendency of the user corresponding to the account through the account portrait information, so as to provide corresponding services for the user corresponding to the account. For example, if the account portrait information indicates that the user corresponding to the account is interested in clothing, clothing advertisements or clothing information can be pushed to the account, such as the address of a popular clothing store or a clothing link with high sales and good reputation. If the account portrait information indicates that the user corresponding to the account is interested in celebrity A, the products endorsed by celebrity A, or the film and television works or variety shows participated by celebrity A can be pushed to the account. If the account portrait information indicates that the user corresponding to the account likes to buy things, a link for purchasing products can be pushed to the account. If the account portrait information indicates that the user corresponding to the account likes to surf the Internet, some download entrances of news platforms, video platforms, short video platforms or community interaction platforms can be recommended to the account.

[0048] Obviously, the ability of account portrait information to represent the interest or behavior tendency of the user corresponding to the account is closely related to the service content and service quality of the services provided by the application. Therefore, many applications in related technologies are committed to optimizing account portrait information and improving the quality of account portrait information. Generally speaking, account portrait information can be obtained by collecting information about the user in various dimensions and fusing this information, and the quality of account portrait information can be improved by information fusion and diversification of information dimensions. However, the fusion method in related technologies is relatively rough, which also affects the quality of account portrait information.

[0049] For example, the relevant technology can perform statistical fusion on the labels in the same account portrait information based on statistics. The fusion dimensions can include time, frequency, and behavior type, so as to construct a fusion formula of three dimensions. The fusion formula can be to set the weights of different dimensions based on experience, so as to generate a label weight pair of <label, weight>. Each label in the account portrait information is configured with a corresponding weight, so as to make the account portrait information more complete. However, this fusion method has the following problems:

[0050] First, the weights in the fusion formula rely more on manual experience and are not universal or referenceable.

[0051] Second, the labels in the account portrait information may come from different data sources, which involves the fusion of multiple data sources. The fusion formula of the related technology does not take into account the role of the variable of different data sources in the fusion, but directly ignores this variable for direct fusion, which is likely to result in information loss. This loss can be reflected in the fusion based on the time dimension or in the fusion based on the behavior type. For example, searching for "A model mobile phone" in data source 1 and reading "A model mobile phone" in data source 2, the amount of information carried by the two types of behaviors is different, and the user intention is also different. Generally speaking, search behavior can better reflect the user's real needs. However, if this difference is not taken into account when fusion is performed, the weight setting may not be accurate enough, resulting in information loss.

[0052] Third, in the process of generating account portrait information, the specific scenarios in which the account portrait information is used to provide related services are not taken into consideration. In other words, the a posteriori indication of the application performance of subsequent applications of the account portrait information on the account portrait information is not considered, making it impossible to optimize the account portrait information for its corresponding subsequent applications, thereby making it impossible for subsequent applications to have better application effects. To a certain extent, this shows that the account portrait information cannot be adaptively changed according to different subsequent applications, which affects the use effect of the account portrait information.

[0053] In summary, the quality of account portrait information in the relevant technology needs to be improved, and there is still room for further optimization and progress. Therefore, the embodiment of the present application provides a method for generating account portrait information, which can effectively integrate the information under the time dimension, behavior dimension and data source dimension in the account behavior record to obtain account portrait information. And in the process of fusion, objective and universal fusion weights are set for the information under these three dimensions, so as to improve the quality of the generated account portrait information. In addition, by improving the setting method of the weight, the dependence on manual work is reduced, so that the generated account portrait information has universality, that is, in different business scenarios, related services based on the account portrait information can achieve good results. The account portrait information generation method also takes into account the information under the data source dimension, so that the factor of the data source is also included in the considerations when generating the account portrait information, further improving the quality of the account portrait information.

[0054] Moreover, in some implementations, the account portrait information generation method may further optimize the quality of the account portrait information by flattening and combining a posteriori data.

[0055] In order to make the objectives, technical solutions and advantages of the present application clearer, the implementation methods of the present application will be further described in detail below in conjunction with the accompanying drawings.

[0056] Please refer to Figure 1 , which shows a schematic diagram of an application program running environment related to a method provided by an embodiment of the present application. The application program running environment may include: a terminal 10 and a server 20.

[0057] The terminal 10 includes, but is not limited to, electronic devices such as mobile phones, computers, intelligent voice interaction devices, smart home appliances, vehicle-mounted terminals, game consoles, e-book readers, multimedia playback devices, wearable devices, etc. The terminal 10 may be installed with a client of an application program.

[0058] In an embodiment of the present application, the above-mentioned application can be any application that can provide an account portrait information generation service. Typically, the application is a video application. Of course, in addition to video applications, other types of applications can also provide information flow content services that rely on account portrait information generation. For example, news applications, social applications, interactive entertainment applications, browser applications, shopping applications, content sharing applications, virtual reality (Virtual Reality, VR) applications, augmented reality (Augmented Reality, AR) applications, etc., are not limited to this in the embodiment of the present application. This is not limited in the embodiment of the present application. Optionally, a client of the above-mentioned application is running in the terminal 10. In some embodiments, the above-mentioned information flow content service covers many vertical contents such as variety shows, film and television, news, finance, sports, entertainment, games, etc., and users can enjoy many forms of content services such as articles, pictures, small videos, short videos, live broadcasts, special topics, columns, etc. through the above-mentioned information flow content service.

[0059] The server 20 is used to provide background services for the client of the application in the terminal 10. For example, the server 20 can be the background server of the above application. The server 20 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or 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, CDN (Content Delivery Network), and big data and artificial intelligence platforms. Optionally, the server 20 provides background services for applications in multiple terminals 10 at the same time.

[0060] Optionally, the terminal 10 and the server 20 may communicate with each other via a network 30. The terminal 10 and the server 20 may be directly or indirectly connected via wired or wireless communication, which is not limited in the present application.

[0061] Please refer to Figure 2 , which shows a flowchart of a method for generating account portrait information provided by an embodiment of the present application. The method can be applied to a computer device, which refers to an electronic device with data calculation and processing capabilities. For example, the execution subject of each step can be Figure 1 The server 20 in the application running environment shown. The method may include the following steps:

[0062] Step 201, obtain account behavior records related to the account, the above account behavior records include a time dimension tag, a behavior dimension tag and a data source dimension tag, the above behavior dimension tag indicates the category associated with the behavior performed by the corresponding account, and the above data source dimension tag indicates the source of the behavior performed by the corresponding account.

[0063] The embodiments of the present application do not limit the source of the account behavior records. For example, when a user visits a website or purchases goods through an application, an account behavior record corresponding to the user's account can be generated on the corresponding website or application. Alternatively, the account behavior record can be extracted from the user's relevant information records with the user's authorization. The embodiments of the present application emphasize that any data related to the user or the user's account used must be authorized by the user. The data acquisition process of the embodiments of the present application is legal and compliant.

[0064] In the Internet environment, users can perform various actions in various time windows in various data sources, thereby generating various account behavior records. In the embodiment of the present application, the time dimension label, behavior dimension label and data source dimension label are used to reflect the information of the account behavior record in the time dimension, behavior dimension and data source dimension respectively.

[0065] The above data source dimension label indicates the source of the behavior performed by the corresponding account. In many cases, users can generate account behaviors in data sources corresponding to different businesses such as e-commerce, retail, and reading. The above behavior dimension indicates the category of specific behavior and the category that bears the behavior. This category can also be considered as the category associated with the corresponding account. Specifically, the category of behavior can be search, collection, attention, click, and other behaviors triggered by users in different business scenarios. The time dimension refers to the different behaviors that users will generate in different time windows. For example, real-time interest is the interest generated by behavior within 1 hour, short-term interest is the interest generated by user behavior within 3 days, medium-term interest refers to the interest generated by users within more than 3 days and less than 1 month, and long-term interest is the interest that is more than 1 month and is classified as long-term interest. Correspondingly, the information corresponding to time windows such as within 1 hour, within 3 days, from 3 days to 1 month, and more than one month are all information under the time dimension.

[0066] Please refer to Figure 3 , which shows a schematic diagram of an account behavior record space provided by an embodiment of the present application. The space formed by the account behavior record can be represented by a three-dimensional coordinate space, and the three dimensions are data source dimension, behavior dimension and time dimension. Figure 3 , multiple behavior records in the three-dimensional coordinate space are shown, and each account behavior record can correspond to a certain point in this three-dimensional coordinate space. For example, "applet A bought fruit an hour ago", "e-commerce B bought a watch a month ago", "watched travel advice a day ago", etc.

[0067] S202. Determine the first weight corresponding to the above-mentioned time dimension tag based on the time distance between the time indicated by the above-mentioned time dimension tag and the time when the account portrait information is generated.

[0068] The embodiment of the present application points out that the above-mentioned first weight is inversely correlated with the above-mentioned time distance; the time indicated by the label corresponding to the above-mentioned account behavior record in the above-mentioned time dimension and the time distance between the generation time of the account portrait information will affect the validity of the information in the account behavior record. The closer the time distance, the more valid the information. The embodiment of the present application does not limit the specific method for determining the first weight, as long as the inverse correlation relationship is satisfied.

[0069] In one feasible implementation, the formula To determine the first weight w time , where action time is the time indicated by the above time dimension label, current time is the time when the account portrait information is generated; ω is a custom weight that can be adjusted according to the portrait effect. The absolute value of the time indicated by the above time dimension label in the account behavior record minus the time when the account portrait information is generated represents the time distance from the current time, which is inversely correlated with the effectiveness of the behavior. The closer to the current time, the greater the weight value of the first weight.

[0070] S203. Determine the second weight corresponding to the above-mentioned behavior dimension label according to the significance index corresponding to the category pointed to by the above-mentioned behavior dimension label.

[0071] In an embodiment of the present application, the weight associated with the category can be determined directly based on the significance index corresponding to the category pointed to by the above-mentioned behavior dimension label, and the weight can be directly used as the second weight corresponding to the above-mentioned behavior dimension label. Different behavior weights can also be set for different behavior types, and the product of the behavior weight and the category-related weight can be used as the second weight. The embodiment of the present application does not limit the method for setting the behavior weight, which can be set according to manual experience. If the main consideration is the impact of the category on the weight, the weight of each behavior can also be set to 1.

[0072] The significance index corresponding to the above-mentioned categories is characterized by the target group index associated with the above-mentioned categories. The category-related weight is inversely correlated with the above-mentioned target group index, so the above-mentioned second weight is inversely correlated with the above-mentioned target group index. The target group index is the TGI index, which is used to evaluate the popularity of the category. If it is a popular category, it means that the category does not have a good degree of differentiation, and the TGI coefficient at this time will be close to 1. Then the weight of such categories tends to be set to the minimum. In other words, the present application tends to restrict the category-related weights corresponding to the popular categories, but does not limit the specific method of restriction. The category-related weights can be manually set to directly restrict, or the popular categories can be downgraded by dividing the preset value by the frequency of the category appearing in the sample, thereby weakening the influence of popular categories on the account portrait information with strong differentiation.

[0073] In one embodiment, the formula Calculate the significance index TGI_category corresponding to the category. ctr_category refers to the average click-through rate of users for this category. ctr_overall_category refers to the average overall click-through rate of the entire category for people who have clicked on this category, thereby evaluating the popularity of the category. For example, if the category is "T-shirts", the average click-through rate of ctr_category (user group for this category) is 0.4, and the click-through rate of people who have clicked on this category (T-shirts) for all categories is also 0.4, indicating that this category is not enough to distinguish this wave of people clicking on T-shirts. Because the click-through rate of all people clicking on this category is also 0.4, which is the same, therefore, the TGI_category for "T-shirts" is 1, and its category weight is set to the minimum.

[0074] S204. Determine a third weight corresponding to the data source dimension tag according to the content of the data source dimension tag.

[0075] The embodiment of the present application may also set corresponding third weights for different data sources. The embodiment of the present application does not limit the method for setting the third weight, which may be set based on manual experience or in combination with the directionality of the specific business corresponding to the data source.

[0076] S205. Generate the account portrait information based on the account behavior record, the first weight, the second weight and the third weight.

[0077] The embodiment of the present application can fuse the above-mentioned account behavior records, the above-mentioned first weight, the above-mentioned second weight and the above-mentioned third weight to generate the above-mentioned account portrait information. The embodiment of the present application does not limit the specific fusion method. The embodiment of the present application provides an account portrait information generation method, which can effectively fuse the information under the time dimension, behavior dimension and data source dimension in the account behavior record to obtain account portrait information. And in the process of fusion, objective and universal fusion weights are set for the information under these three dimensions, so as to improve the quality of the generated account portrait information. In addition, by improving the setting method of the weight, the dependence on manual work is reduced, so that the generated account portrait information has universality, that is, in different business scenarios, providing related services based on the account portrait information can achieve good results. The account portrait information generation method also takes into account the information under the data source dimension, so that the factor of the data source is also included in the consideration factors when generating the account portrait information, further improving the quality of the account portrait information.

[0078] In one embodiment, the account portrait information includes an account-category portrait, which indicates the interest tendency of the account in the above-mentioned category; the account portrait information is generated according to the account behavior record, the first weight, the second weight and the third weight, including: fusing the first weight, the second weight and the third weight to obtain the fusion weight of the account for the above-mentioned category, and the fusion weight indicates the interest tendency degree of the account in the above-mentioned category. The embodiment of the present application directly fuses these three weights, but does not limit the specific fusion method. For example, the three weights can be directly multiplied to obtain the fusion weight. The larger the weight value of the fusion weight, the more interested the user is in this category.

[0079] Then, the target category system can be obtained. The target category system is a category system common to all data sources. According to the target category system and the account behavior records, account behavior entries are generated. This embodiment of the application sets a unified category system for different data sources to better and more standardizedly integrate information. The target category system proposed in this embodiment of the application includes four-level categories, brands, and product terms. For details, please refer to Figure 4 , which shows a schematic diagram of an account-category portrait provided in an embodiment of the present application. The account-category portrait is generated by the above account behavior items and the above fusion weights. Figure 4 On the left are unified categories (first level, second level, third level, fourth level), brands and product words, and on the right are the fusion weights.

[0080] In other embodiments, the account portrait information may be generated in other ways. Of course, the account portrait information generated in other embodiments and the account-category portrait may also form account portrait information with more information.

[0081] In one embodiment, please refer to Figure 5 , which shows a schematic diagram of another method for generating account portrait information according to an embodiment of the present application. The method obtains account portrait information by fusing the category dimension label, time dimension label and data source dimension label included in the above account behavior record. The method specifically includes:

[0082] S501. Perform tag-based information expansion on the above account behavior records to obtain multiple target tags;

[0083] The so-called tag-based information expansion in the embodiment of the present application can be understood as one of the steps in a flat fusion of the information carried by each tag in the account behavior record. This fusion method can reduce information loss. For example, the amount of information covered by a user buying a watch at e-commerce A one month ago and buying pants at e-commerce A one day ago is different. The amount of information covered by a user watching travel advice one day ago and buying fruit one hour ago is also different. That is, when generating account portrait information, it is not recommended to simply fuse the weights corresponding to each dimension, such as multiplying them. A better way is to pay full attention to the different information loads of information under each dimension, reduce information loss through flat fusion, and improve the quality of account portrait information.

[0084] The first step of flat fusion is to expand the feature load, which is achieved through label-based information expansion. Specifically, the features (labels) can be expanded in the time dimension, data source dimension, and behavior dimension. Please refer to Figure 6 , which shows a schematic diagram of the information expansion method in flat fusion. The embodiment of the present application not only retains the information of the time dimension, data source dimension, and behavior dimension, but also performs a second-order cross-combination or a third-order cross-combination on their related labels to generate more labels. Specifically, retain the time dimension label and the data source dimension label, and fuse them on the behavior dimension label; retain the time dimension label and the behavior dimension label, and fuse them on the data source dimension label; retain the data source dimension label and the behavior dimension label, and fuse them on the time dimension label; or a third-order cross-combination. The three dimensions of time dimension label, data source dimension label, and behavior dimension label are all retained.

[0085] Of course, there can also be first-order information; retain the data source dimension label, and fuse it with the behavior dimension label and the time dimension label; retain the time dimension label, and fuse it with the behavior dimension label and the data source dimension label; retain the behavior dimension label, and fuse it with the data source dimension label and the time dimension label.

[0086] Of course, different time dimension labels are related to the time dimension windows divided according to different businesses. For example, time can be divided into 1 hour, 12 hours, 1 day and other time dimension windows. Different time dimension windows can be set according to different business backgrounds. In this way, for example, the label in the account behavior record of the book purchased on e-commerce A the day before can be expanded to multiple target labels such as book, 1day_book, e-commerce A_book, 1day_e-commerce A_book, etc., thereby significantly increasing the feature space volume. For details, please refer to Figure 7 , which shows a schematic diagram of the feature expansion effect. The embodiment of the present application can directly expand the feature space corresponding to the original label of the account behavior record from about 50,000 dimensions to 14 million dimensions.

[0087] In an exemplary embodiment, the target tag includes at least one of the following:

[0088] A single-order label formed by a category dimension label, a time dimension label, or a data source dimension label, where the above-mentioned category dimension label belongs to the above-mentioned behavior dimension label;

[0089] A second-order label is obtained by performing a second-order cross combination based on the labels of any two dimensions among the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label;

[0090] A third-order label is obtained based on a third-order cross combination of the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label.

[0091] For example, if a user buys a science fiction book "Book Title A", the target category system mapped by Book Title A is:

[0092] The first-level category is books.

[0093] The secondary category is technology.

[0094] The third-level category is popular science books.

[0095] Brand is Chinese science fiction,

[0096] The product words are the book title A, the author B, and the universe.

[0097] This corresponds to the target category system mentioned above.

[0098] In this way, the user's behavior of buying a book title A on e-commerce A one day ago can be mapped to an account behavior record, where the complete characteristics of the category in the behavior dimension label include "Books - Technology - Popular Science Books - Chinese Science Fiction Novels - Book Title A's Author B; Universe", and the behavior type is purchase. The time dimension label is 1 day ago, and the data source dimension label is e-commerce A. By performing label-based information expansion on the account behavior record, the following target labels can be obtained:

[0099] 1) Only categories

[0100] Books, Books-Technology, Books-Technology-Popular Science Books, Books-Technology-Popular Science Books-Chinese Science Fiction Novels-Author B of Title A; Universe

[0101] 2) Category + time dimension label

[0102] 1day_books, 1day_books-technology, 1day_books-technology-popular science books, 1day_books-technology-popular science books-Chinese science fiction-title A's author B; universe

[0103] 3) Category + behavior type

[0104] buy_books, buy_books-technology, buy_books-technology-popular science books, buy_books-technology-popular science books-Chinese science fiction novels-author B of title A; universe

[0105] 4) Category + data source dimension label

[0106] E-commerce A_Books, E-commerce A_Books-Technology, E-commerce A_Books-Technology-Popular Science Books, E-commerce A_Books-Technology-Popular Science Books-Chinese Science Fiction Novels-Author B of Title A; Universe

[0107] 5) Category + behavior type + time dimension label

[0108] 1day_buy_books, 1day_buy_books-technology, 1day_buy_books-technology-popular science books, 1day_buy_books-technology-popular science books-Chinese science fiction novels-author B of title A; universe

[0109] 6) Category + time dimension label + data source dimension label

[0110] 7) Category + behavior type + time dimension label + data source dimension label

[0111] It can be seen that by performing tag-based information expansion on the account behavior record, more target tags can be obtained, which can significantly increase the volume of tags and the information carrying capacity.

[0112] S502. Calculate the target weight corresponding to each of the target tags according to the first weight, the second weight and the third weight.

[0113] In the embodiment of the present application, the target weight calculation method of the corresponding target tag can be obtained according to different information expansion methods. Specifically, the target weight corresponding to each of the above target tags is calculated according to the above first weight, the above second weight and the above third weight, including: when the above target tag is a single-order tag, the weight corresponding to the above single-order tag is used as the above target weight; when the above target tag is a second-order tag, the weights corresponding to the two-dimensional tags corresponding to the above target tag are respectively fused to obtain the above target weight; when the above target tag is a third-order tag, the weights corresponding to the three-dimensional tags corresponding to the above target tag are respectively fused to obtain the above target weight.

[0114] The present application does not limit the method of fusing different weights. For example, they can be directly multiplied or weighted added. Of course, the weights can be set based on experience.

[0115] For example, for a second-order target label, it corresponds to a time dimension label and a data source dimension label, and the time dimension label and the data source dimension label correspond to the first weight and the third weight respectively, then the result of 0.4*first weight+0.6*third weight can be used as the target weight of the target label. For a third-order target label, the result of 0.4*first weight+0.3*second weight+0.3*third weight can be used as the target weight of the target label.

[0116] S503. Generate the account portrait information based on each of the above target tags and the target weight corresponding to each of the above target tags.

[0117] In the embodiment of the present application, each of the above target tags and the target weight corresponding to each of the above target tags can be used as the content in the account portrait information, so as to obtain the account portrait information. The embodiment of the present application introduces the "flat fusion" method to cross-fuse the information of the three dimensions of time dimension, data source dimension and behavior dimension, expand the features, reduce information loss, and improve the quality of account portrait information.

[0118] The process of constructing account portrait information in related technologies is usually not directly associated with the posterior effect of the account portrait information when it is applied. Usually, after the account portrait information is generated, the account portrait information is synchronized to the relevant model for subsequent application, or synchronized to the strategy team of the subsequent application for application, and then the application results are fed back to the portrait generation team, and then the portrait generation team adjusts the content of the account portrait information based on experience. The problem in this case is that in the process of constructing the account portrait information, its adaptability to subsequent applications is not directly considered, which leads to unsatisfactory performance in subsequent applications based on the account portrait information.

[0119] In one embodiment, when constructing the account portrait information, the effect of the account portrait information on the subsequent application link can be directly considered, and the content of the account portrait information can be adjusted based on the feedback of the effect. Specifically, based on the a posteriori data related to the above account portrait information, a behavior sample corresponding to the a posteriori behavior can be generated, and the above behavior sample includes an account identifier with the above account portrait information, the above account portrait information, and the a posteriori behavior generated based on the above account portrait information; based on the above behavior sample, the evaluation index value corresponding to the above account portrait information is calculated, and the above evaluation index value is used to indicate the use effect of the above account portrait information in the application scenario related to the above a posteriori behavior; according to the above evaluation index value, the relevant weights in the above account portrait information are adjusted. The a posteriori behavior in the embodiment of the present application refers to the relevant behavior corresponding to the aforementioned subsequent application, and the a posteriori data is the true performance of these a posteriori behaviors in the actual scenario.

[0120] The embodiments of this application do not limit the scenarios of subsequent applications. The inventive concept is consistent. The following is just an example and does not constitute a limitation of the embodiments of this application. Taking the advertising recommendation scenario as an example, the purpose of the embodiments of this application in the advertising recommendation scenario is to improve the real click-through rate of advertisements by optimizing the quality of account portrait information.

[0121] In the advertising recommendation scenario, the above-mentioned a posteriori behavior is the exposure or click behavior in the advertising recommendation scenario. The above-mentioned a posteriori data related to the above-mentioned account portrait information generates a behavior sample corresponding to the a posteriori behavior, including: obtaining the account ID of the account with the above-mentioned account portrait information; obtaining the advertising ID of the advertisement pushed to the above-mentioned account ID, and the a posteriori behavior generated by the above-mentioned account ID for the above-mentioned advertising ID, wherein the push time of the above-mentioned advertisement is later than the generation time of the above-mentioned account portrait information; constructing the above-mentioned behavior sample according to the above-mentioned account ID, the above-mentioned account portrait information, the above-mentioned advertising ID and the above-mentioned a posteriori behavior.

[0122] First, we can get <user id, feature>, where user id is the account ID of the account with the above account portrait information, and feature refers to the account portrait information. Then, we can get the advertising behavior, which is <user id, advertising id, behavior (click / exposure)>, and advertising id is the advertising ID of the advertisement pushed to the above account ID. In this way, we can get the behavior sample <user id, user portrait feature, advertising id, behavior (exposure / click)>.

[0123] The purpose of limiting the push time of the above-mentioned advertisement to later than the generation time of the above-mentioned account portrait information in the embodiment of the present application is to avoid feature crossing. Feature crossing will cause the advertising recommendation model to use future features when recommending advertisements based on account portrait information, which will lead to a high bias in advertising prediction. The embodiment of the present application does not limit the advertising recommendation model, which can be modeled based on deep learning and machine learning methods. Specifically, it can be built and trained using the BERT model in the relevant technology or the related structure of Transformer.

[0124] Please refer to Figure 8 , which shows a schematic diagram of the timing requirements for constructing behavior samples. The account portrait information (user interests) generated at 15 can be combined with the advertising exposure behavior at 16 to generate a behavior sample. In other words, the correct order is that the account portrait information is generated first and the advertising exposure behavior is followed. The construction of behavior samples should follow this correct order.

[0125] In one embodiment, the above-mentioned calculation of the evaluation index value corresponding to the above-mentioned account portrait information based on the above-mentioned behavior sample includes: performing data statistics on the above-mentioned behavior sample to obtain the number of clicks corresponding to the above-mentioned account portrait information and the number of exposures corresponding to the above-mentioned account portrait information; calculating the click-through rate corresponding to the above-mentioned account portrait information based on the ratio of the number of clicks corresponding to the above-mentioned account portrait information to the number of exposures corresponding to the above-mentioned account portrait information, and the above-mentioned click-through rate indicates the above-mentioned evaluation index value. For example, according to the formula Calculate the click-through rate, that is, the indicator evaluation value, where clk_feature is the number of clicks corresponding to the account portrait information, which is the true value obtained based on the posterior data, and imp_feature is the amount of advertising exposure corresponding to the account portrait information, which is also the true value obtained based on the posterior data, thereby calculating the true click-through rate corresponding to the account portrait information, that is, ctr_feature.

[0126] In one embodiment, the above-mentioned adjusting the relevant weights in the above-mentioned account portrait information according to the above-mentioned evaluation index value includes: normalizing the above-mentioned click-through rate to obtain updated weights; and fusing the above-mentioned updated weights with the relevant weights in the above-mentioned account portrait information to obtain updated relevant weights.

[0127] The so-called relevant weights in the embodiments of the present application are the various weights in the account portrait information, that is, the various weights in the account portrait information can be adjusted based on the update weights. The embodiments of the present application do not limit the way of fusing the above-mentioned update weights with the relevant weights in the above-mentioned account portrait information, for example, directly multiplying them. Specifically, the purpose of the normalization process is to obtain an update weight between 0 and 1 to update the various weights in the account portrait information. The normalization formula is as follows:

[0128]

[0129] Where ctr_max is the maximum ctr value that can be obtained in each account portrait information; ctr_min is the minimum ctr value that can be obtained in each account portrait information; ctr_feature is the ctr value under the aforementioned account portrait information. The ctr value is the value of CTR, and CTR (Click-Through-Rate) is a commonly used term in Internet advertising, which is not elaborated in this embodiment of the application.

[0130] The embodiment of the present application uses the a posteriori data generated on the a posteriori application link to perform statistical analysis, and directly applies the results of the statistical analysis to the values ​​of the relevant weights in the account portrait information, thereby updating the account portrait information content and improving the quality of the account portrait information, thereby improving its implementation effect on the application link. Taking the effect of the technical solution of the embodiment of the present application in the field of video number advertising recommendation applications as an example, the account portrait information obtained by applying the embodiment of the present application can significantly increase CTR by 0.31% and GMV by 1.33%. GMV is a concept in e-commerce operations. It is the acronym for Gross Merchandise Volume, which means the total amount of commercial transactions, and is used to indicate the total transaction amount over a period of time.

[0131] The following is an embodiment of the device of the present application, which can be used to execute the embodiment of the method of the present application. For details not disclosed in the embodiment of the device of the present application, please refer to the embodiment of the method of the present application.

[0132] Please refer to Fig. 9 , which shows a block diagram of an account portrait information generation device provided by an embodiment of the present application. The device has the function of implementing the above-mentioned account portrait information generation method, and the above-mentioned function can be implemented by hardware, or by hardware executing corresponding software. The device can be a computer device, or can be set in a computer device. The device may include:

[0133] The account data extraction module 901 is used to obtain account behavior records related to the account. The account behavior records include a time dimension tag, a behavior dimension tag, and a data source dimension tag. The behavior dimension tag indicates the category associated with the behavior performed by the corresponding account, and the data source dimension tag indicates the source of the behavior performed by the corresponding account.

[0134] The account profile information generation module 902 is used to perform the following operations:

[0135] Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account profile information is generated;

[0136] Determine a second weight corresponding to the behavior dimension label according to the significance index corresponding to the category pointed to by the behavior dimension label;

[0137] Determine, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag;

[0138] The account portrait information is generated based on the account behavior record, the first weight, the second weight and the third weight.

[0139] In one embodiment, the account profile information generation module 902 is used to perform the following operations:

[0140] Perform tag-based information expansion on the above account behavior records to obtain multiple target tags;

[0141] Calculate the target weight corresponding to each of the target tags according to the first weight, the second weight and the third weight;

[0142] The account portrait information is generated based on each of the above target tags and the target weight corresponding to each of the above target tags.

[0143] In one embodiment, the account profile information generation module 902 is used to perform the following operations:

[0144] Based on the a posteriori data related to the account portrait information, a behavior sample corresponding to the a posteriori behavior is generated, wherein the behavior sample includes an account identifier having the account portrait information, the account portrait information, and the a posteriori behavior generated based on the account portrait information;

[0145] Based on the above behavior samples, the evaluation index value corresponding to the above account portrait information is calculated, and the above evaluation index value is used to indicate the use effect of the above account portrait information in the application scenario related to the above posterior behavior;

[0146] Adjust the relevant weights in the above account portrait information according to the above evaluation index values.

[0147] In one embodiment, the first weight is inversely correlated with the time distance;

[0148] The significance index corresponding to the above categories is represented by the target group index associated with the above categories, and the above second weight is inversely correlated with the above target group index.

[0149] In one embodiment, the target tag includes at least one of the following:

[0150] A single-order label formed by a category dimension label, a time dimension label, or a data source dimension label, where the above-mentioned category dimension label belongs to the above-mentioned behavior dimension label;

[0151] A second-order label is obtained by performing a second-order cross combination based on the labels of any two dimensions among the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label;

[0152] A third-order label is obtained based on a third-order cross combination of the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label.

[0153] In one embodiment, the account profile information generation module 902 is used to perform the following operations:

[0154] When the target label is a single-order label, the weight corresponding to the single-order label is used as the target weight;

[0155] When the target label is a second-order label, the weights corresponding to the labels of the two dimensions corresponding to the third-order target label are respectively fused to obtain the target weight;

[0156] When the target label is a three-dimensional label, the weights corresponding to the labels of the three dimensions corresponding to the target label are fused to obtain the target weight.

[0157] In one embodiment, the a posteriori behavior is an exposure or click behavior in an advertisement recommendation scenario, and the account portrait information generation module 902 is used to perform the following operations:

[0158] Obtain the account ID of the account having the above account profile information;

[0159] Obtaining the advertisement identifier of the advertisement pushed to the above-mentioned account identifier, and the a posteriori behavior generated by the above-mentioned account identifier for the above-mentioned advertisement identifier, wherein the push time of the above-mentioned advertisement is later than the generation time of the above-mentioned account portrait information;

[0160] The above-mentioned behavior sample is constructed based on the above-mentioned account identification, the above-mentioned account portrait information, the above-mentioned advertising identification and the above-mentioned posterior behavior.

[0161] In one embodiment, the account profile information generation module 902 is used to perform the following operations:

[0162] Performing data statistics on the above behavior samples to obtain the number of clicks corresponding to the above account portrait information and the number of exposures corresponding to the above account portrait information;

[0163] According to the ratio of the number of clicks corresponding to the above account portrait information and the number of exposures corresponding to the above account portrait information, the click rate corresponding to the above account portrait information is calculated, and the above click rate indicates the above evaluation index value.

[0164] In one embodiment, the account profile information generation module 902 is used to perform the following operations:

[0165] The above click rates are normalized to obtain updated weights;

[0166] The above-mentioned updated weights are merged with the relevant weights in the above-mentioned account portrait information to obtain updated relevant weights.

[0167] In one embodiment, the account portrait information includes an account-category portrait, and the account-category portrait indicates the interest tendency of the account in the category; the account portrait information generating module 902 is used to perform the following operations:

[0168] The first weight, the second weight and the third weight are integrated to obtain an integrated weight of the account for the category, where the integrated weight indicates the interest tendency of the account for the category;

[0169] Obtain a target category system, where the target category system is a category system common to all data sources;

[0170] Generate account behavior entries based on the above target category system and the above account behavior records;

[0171] Based on the above account behavior items and the above fusion weights, the above account-category portrait is obtained.

[0172] It should be noted that the device provided in the above embodiment, when implementing its functions, is only illustrated by the division of the above functional modules. In actual applications, the above functions can be assigned to different functional modules as needed, that is, the internal structure of the device is divided into different functional modules to complete all or part of the functions described above. In addition, the device and method embodiments provided in the above embodiment belong to the same concept, and the specific implementation process is detailed in the method embodiment, which will not be repeated here.

[0173] Please refer to Fig.10 , which shows a block diagram of a computer device provided by an embodiment of the present application. The computer device may be a server for executing the above-mentioned account portrait information generation method. Specifically:

[0174] The computer device 1000 includes a central processing unit (CPU) 1001, a system memory 1004 including a random access memory (RAM) 1002 and a read only memory (ROM) 1003, and a system bus 1005 connecting the system memory 1004 and the central processing unit 1001. The computer device 1000 also includes a basic input / output system (I / O (Input / Output) system) 1006 for facilitating information transmission between various components in the computer, and a large-capacity storage device 1007 for storing an operating system 1013, application programs 1014, and other program modules 1015.

[0175] The basic input / output system 1006 includes a display 1008 for displaying information and an input device 1009 such as a mouse and a keyboard for user inputting information. The display 1008 and the input device 1009 are connected to the central processing unit 1001 through an input / output controller 1010 connected to the system bus 1005. The basic input / output system 1006 may also include an input / output controller 1010 for receiving and processing inputs from a plurality of other devices such as a keyboard, a mouse, or an electronic stylus. Similarly, the input / output controller 1010 also provides output to a display screen, a printer, or other types of output devices.

[0176] The mass storage device 1007 is connected to the central processing unit 1001 through a mass storage controller (not shown) connected to the system bus 1005. The mass storage device 1007 and its associated computer readable medium provide non-volatile storage for the computer device 1000. That is, the mass storage device 1007 may include a computer readable medium (not shown) such as a hard disk or a CD-ROM (Compact Disc Read-Only Memory) drive.

[0177] Without loss of generality, computer readable media may include computer storage media and communication media. Computer storage media include volatile and non-volatile, removable and non-removable media implemented by any method or technology for storing information such as computer readable instructions, data structures, program modules or other data. Computer storage media include RAM, ROM, EPROM (Erasable Programmable Read Only Memory), EEPROM (Electrically Erasable Programmable Read Only Memory), flash memory or other solid-state storage technologies, CD-ROM, DVD (Digital Video Disc) or other optical storage, cassettes, magnetic tapes, disk storage or other magnetic storage devices. Of course, those skilled in the art will appreciate that computer storage media are not limited to the above. The above-mentioned system memory 1004 and mass storage device 1007 can be collectively referred to as memory.

[0178] According to various embodiments of the present application, the computer device 1000 can also be connected to a remote computer on the network through a network such as the Internet. That is, the computer device 1000 can be connected to the network 1012 through the network interface unit 1011 connected to the system bus 1005, or the network interface unit 1011 can be used to connect to other types of networks or remote computer systems (not shown).

[0179] The above-mentioned memory also includes a computer program, which is stored in the memory and configured to be executed by one or more processors to implement the above-mentioned account portrait information generation method.

[0180] In an exemplary embodiment, a computer-readable storage medium is also provided, in which at least one instruction, at least one program, a code set or an instruction set is stored. When the at least one instruction, the at least one program, the code set or the instruction set is executed by a processor, the account portrait information generation method is implemented.

[0181] Specifically, the account portrait information generation method includes:

[0182] Obtaining account behavior records related to the account, the account behavior records including a time dimension tag, a behavior dimension tag, and a data source dimension tag, the behavior dimension tag indicating a category associated with the behavior performed by the corresponding account, and the data source dimension tag indicating a source of the behavior performed by the corresponding account;

[0183] Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account profile information is generated;

[0184] Determine a second weight corresponding to the behavior dimension label according to the significance index corresponding to the category pointed to by the behavior dimension label;

[0185] Determine, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag;

[0186] The account portrait information is generated based on the account behavior record, the first weight, the second weight and the third weight.

[0187] In one embodiment, generating the account profile information according to the account behavior record, the first weight, the second weight and the third weight includes:

[0188] Perform tag-based information expansion on the above account behavior records to obtain multiple target tags;

[0189] Calculate the target weight corresponding to each of the target tags according to the first weight, the second weight and the third weight;

[0190] The account portrait information is generated based on each of the above target tags and the target weight corresponding to each of the above target tags.

[0191] In one embodiment, the method further includes:

[0192] Based on the a posteriori data related to the account portrait information, a behavior sample corresponding to the a posteriori behavior is generated, wherein the behavior sample includes an account identifier having the account portrait information, the account portrait information, and the a posteriori behavior generated based on the account portrait information;

[0193] Based on the above behavior samples, the evaluation index value corresponding to the above account portrait information is calculated, and the above evaluation index value is used to indicate the use effect of the above account portrait information in the application scenario related to the above posterior behavior;

[0194] Adjust the relevant weights in the above account portrait information according to the above evaluation index values.

[0195] In one embodiment, the first weight is inversely correlated with the time distance;

[0196] The significance index corresponding to the above categories is represented by the target group index associated with the above categories, and the above second weight is inversely correlated with the above target group index.

[0197] In one embodiment, the target tag includes at least one of the following:

[0198] A single-order label formed by a category dimension label, a time dimension label, or a data source dimension label, where the above-mentioned category dimension label belongs to the above-mentioned behavior dimension label;

[0199] A second-order label is obtained by performing a second-order cross combination based on the labels of any two dimensions among the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label;

[0200] A third-order label is obtained based on a third-order cross combination of the above-mentioned category dimension label, the above-mentioned time dimension label and the above-mentioned data source dimension label.

[0201] In one embodiment, calculating the target weight corresponding to each target tag according to the first weight, the second weight and the third weight includes:

[0202] When the target label is a single-order label, the weight corresponding to the single-order label is used as the target weight;

[0203] When the target label is a second-order label, the weights corresponding to the labels of the two dimensions corresponding to the target label are fused to obtain the target weight;

[0204] In the case where the target label is a three-level label, the weights corresponding to the labels of the three dimensions corresponding to the target label are fused to obtain the target weight.

[0205] In one embodiment, the a posteriori behavior is an exposure or click behavior in an advertisement recommendation scenario, and the a posteriori data related to the account portrait information is used to generate a behavior sample corresponding to the a posteriori behavior, including:

[0206] Obtain the account ID of the account having the above account profile information;

[0207] Obtaining the advertisement identifier of the advertisement pushed to the above-mentioned account identifier, and the a posteriori behavior generated by the above-mentioned account identifier for the above-mentioned advertisement identifier, wherein the push time of the above-mentioned advertisement is later than the generation time of the above-mentioned account portrait information;

[0208] The above-mentioned behavior sample is constructed based on the above-mentioned account identification, the above-mentioned account portrait information, the above-mentioned advertising identification and the above-mentioned posterior behavior.

[0209] In one embodiment, the calculation of the evaluation index value corresponding to the account profile information based on the behavior sample includes:

[0210] Performing data statistics on the above behavior samples to obtain the number of clicks corresponding to the above account portrait information and the number of exposures corresponding to the above account portrait information;

[0211] According to the ratio of the number of clicks corresponding to the above account portrait information and the number of exposures corresponding to the above account portrait information, the click rate corresponding to the above account portrait information is calculated, and the above click rate indicates the above evaluation index value.

[0212] In one embodiment, adjusting the relevant weights in the account profile information according to the evaluation index value includes:

[0213] The above click rates are normalized to obtain updated weights;

[0214] The above-mentioned updated weights are merged with the relevant weights in the above-mentioned account portrait information to obtain updated relevant weights.

[0215] In one embodiment, the account portrait information includes an account-category portrait, and the account-category portrait indicates the interest tendency of the account in the category; the account portrait information is generated according to the account behavior record, the first weight, the second weight, and the third weight, including:

[0216] The first weight, the second weight and the third weight are integrated to obtain an integrated weight of the account for the category, where the integrated weight indicates the interest tendency of the account for the category;

[0217] Obtain a target category system, where the target category system is a category system common to all data sources;

[0218] Generate account behavior entries based on the above target category system and the above account behavior records;

[0219] Based on the above account behavior items and the above fusion weights, the above account-category portrait is obtained.

[0220] Optionally, the computer-readable storage medium may include: ROM (Read Only Memory), RAM (Random Access Memory), SSD (Solid State Drives) or an optical disk, etc. Among them, the random access memory may include ReRAM (Resistance Random Access Memory) and DRAM (Dynamic Random Access Memory).

[0221] In an exemplary embodiment, a computer program product or a computer program is also provided, the computer program product or the computer program comprising computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of a computer device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the computer device executes the above-mentioned account portrait information generation method.

[0222] It should be understood that the "multiple" mentioned in this article refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. In addition, the step numbers described in this article only illustrate a possible execution sequence between the steps. In some other embodiments, the above steps may not be executed in the order of the numbers, such as two steps with different numbers are executed at the same time, or two steps with different numbers are executed in the opposite order to the diagram. The embodiments of the present application are not limited to this.

[0223] In addition, in the specific implementation of this application, related data such as user information is involved. When the above embodiments of this application are applied to specific products or technologies, user permission or consent is required, and the collection, use and processing of relevant data must comply with relevant laws, regulations and standards of relevant countries and regions.

[0224] The above description is only an exemplary embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present application shall be included in the protection scope of the present application.

Claims

1. A method for generating account portrait information, characterized in that: The method comprises: Acquire account behavior records related to the account, wherein the account behavior records include a time dimension tag, a behavior dimension tag, and a data source dimension tag, wherein the behavior dimension tag indicates a category associated with a behavior performed by the corresponding account, and the data source dimension tag indicates a source of the behavior performed by the corresponding account; Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account portrait information is generated; Determining a second weight corresponding to the behavior dimension label according to a significance index corresponding to the category pointed to by the behavior dimension label; Determining, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag; The account portrait information is generated according to the account behavior record, the first weight, the second weight and the third weight.

2. The method according to claim 1, characterized in that The generating the account portrait information according to the account behavior record, the first weight, the second weight and the third weight includes: Performing tag-based information expansion on the account behavior record to obtain multiple target tags; Calculate the target weight corresponding to each target tag according to the first weight, the second weight and the third weight; The account portrait information is generated according to each of the target tags and the target weight corresponding to each of the target tags.

3. The method according to claim 1 or 2, characterized in that: The method further comprises: Based on the a posteriori data related to the account portrait information, generating a behavior sample corresponding to the a posteriori behavior, the behavior sample including an account identifier having the account portrait information, the account portrait information, and the a posteriori behavior generated based on the account portrait information; Calculate, based on the behavior sample, an evaluation index value corresponding to the account portrait information, wherein the evaluation index value is used to indicate the use effect of the account portrait information in an application scenario related to the posterior behavior; Adjust the relevant weights in the account portrait information according to the evaluation index value.

4. The method according to claim 1, characterized in that: The first weight is inversely correlated with the time distance; The significance index corresponding to the category is represented by a target group index associated with the category, and the second weight is inversely correlated with the target group index.

5. The method according to claim 2, characterized in that: The target tag includes at least one of the following: A single-order label formed by a category dimension label, a time dimension label, or a data source dimension label, wherein the category dimension label belongs to the behavior dimension label; A second-order label obtained by performing a second-order cross combination based on the labels of any two dimensions among the category dimension label, the time dimension label and the data source dimension label; A third-order label is obtained based on a third-order cross combination of the category dimension label, the time dimension label and the data source dimension label.

6. The method according to claim 5, characterized in that The calculating the target weight corresponding to each target tag according to the first weight, the second weight and the third weight includes: When the target label is a single-order label, the weight corresponding to the single-order label is used as the target weight; When the target label is a second-order label, the weights corresponding to the labels of the two dimensions corresponding to the target label are respectively fused to obtain the target weight; When the target label is a three-level label, the weights corresponding to the labels of the three dimensions corresponding to the target label are fused to obtain the target weight.

7. The method according to claim 3, characterized in that The a posteriori behavior is an exposure or click behavior in an advertisement recommendation scenario, and the a posteriori data related to the account portrait information is used to generate a behavior sample corresponding to the a posteriori behavior, including: Obtaining the account ID of the account having the account profile information; Obtaining an advertisement identifier of the advertisement pushed to the account identifier, and a posteriori behavior generated by the account identifier for the advertisement identifier, wherein the push time of the advertisement is later than the generation time of the account portrait information; The behavior sample is constructed according to the account identifier, the account portrait information, the advertisement identifier and the a posteriori behavior.

8. The method according to claim 7, characterized in that The calculating, based on the behavior sample, an evaluation index value corresponding to the account portrait information includes: Performing data statistics on the behavior samples to obtain the number of clicks corresponding to the account portrait information and the number of exposures corresponding to the account portrait information; The click rate corresponding to the account portrait information is calculated based on the ratio of the number of clicks corresponding to the account portrait information and the amount of exposure corresponding to the account portrait information, and the click rate indicates the evaluation index value.

9. The method according to claim 8, characterized in that The adjusting the relevant weights in the account portrait information according to the evaluation index value includes: Normalizing the click rate to obtain an updated weight; The updated weights are merged with the relevant weights in the account portrait information to obtain updated relevant weights.

10. The method according to claim 1, characterized in that The account portrait information includes an account-category portrait, and the account-category portrait indicates the interest tendency of the account in the category; The generating the account portrait information according to the account behavior record, the first weight, the second weight and the third weight includes: fusing the first weight, the second weight, and the third weight to obtain a fusion weight of the account for the category, wherein the fusion weight indicates the interest tendency of the account for the category; Acquire a target category system, where the target category system is a category system common to all data sources; Generate account behavior entries according to the target category system and the account behavior records; The account-category portrait is obtained according to the account behavior items and the fusion weights.

11. An account portrait information generating device, characterized in that: The device comprises: An account data extraction module, used to obtain account behavior records related to an account, wherein the account behavior records include a time dimension tag, a behavior dimension tag, and a data source dimension tag, wherein the behavior dimension tag indicates a category associated with a behavior performed by a corresponding account, and the data source dimension tag indicates a source of generation of the behavior performed by the corresponding account; The account profile information generation module is used to perform the following operations: Determine a first weight corresponding to the time dimension tag according to the time distance between the time indicated by the time dimension tag and the time when the account portrait information is generated; Determining a second weight corresponding to the behavior dimension label according to a significance index corresponding to the category pointed to by the behavior dimension label; Determining, according to the content of the data source dimension tag, a third weight corresponding to the data source dimension tag; The account portrait information is generated according to the account behavior record, the first weight, the second weight and the third weight.

12. A computer device, characterized in that: The computer device includes a processor and a memory, wherein the memory stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the account portrait information generation method as described in any one of claims 1 to 10.

13. A computer-readable storage medium, characterized in that: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, and the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to implement the account portrait information generation method as described in any one of claims 1 to 10.

Images