CN111046275A

CN111046275A - User label determining method and device based on artificial intelligence and storage medium

Info

Publication number: CN111046275A
Application number: CN201911135598.4A
Authority: CN
Inventors: 陈鑫; 闫肃
Original assignee: Tencent Technology Shenzhen Co Ltd
Current assignee: Tencent Technology Shenzhen Co Ltd
Priority date: 2019-11-19
Filing date: 2019-11-19
Publication date: 2020-04-21
Anticipated expiration: 2039-11-19
Also published as: CN111046275B

Abstract

The embodiment of the disclosure provides a user label determining method and device based on artificial intelligence and a storage medium, which relate to the technical field of artificial intelligence and deep learning technology, and the method comprises the following steps: acquiring user data corresponding to a target user in at least one characteristic domain; determining the intra-domain features corresponding to each feature domain according to the user data contained in each feature domain and a plurality of attribute vectors used for representing attribute directions; fusing the intra-domain features to obtain inter-domain features corresponding to the target users, and determining the target features according to the inter-domain features; and identifying the target characteristics to generate a user label of the target user. According to the embodiment of the invention, through the attribute vector, the information loss in the feature extraction process is avoided, and the accuracy of label determination is improved.

Description

User label determining method and device based on artificial intelligence and storage medium

Technical Field

The present disclosure relates to the field of artificial intelligence technology, and in particular, to an artificial intelligence-based user tag determination method, an artificial intelligence-based user tag determination apparatus, and a computer-readable storage medium.

Background

With the development of artificial intelligence technology, it is the key point of information recommendation to accurately recommend information of interest to users.

In the related art, a user representation and thus a tag of information of interest is generally determined by user behavior. However, in the related art, the user tags are generally determined by fusing at the upper layer of the model, which has certain limitations and may cause information loss, so that the accuracy of the determined tags is poor.

In view of the above, there is a need in the art to develop a new method for determining a user tag based on artificial intelligence.

It is to be noted that the information disclosed in the above background section is only for enhancement of understanding of the background of the present application and therefore may include information that does not constitute prior art known to a person of ordinary skill in the art.

Disclosure of Invention

The embodiment of the disclosure provides a user label determining method based on artificial intelligence, a user label determining device based on artificial intelligence and a computer readable storage medium, so that information loss can be avoided at least to a certain extent, an accurate user label is obtained according to comprehensive user data, and the label determining accuracy is improved.

Additional features and advantages of the disclosure will be set forth in the detailed description which follows, or in part will be obvious from the description, or may be learned by practice of the disclosure.

According to an aspect of the embodiments of the present disclosure, there is provided a method for determining a user tag based on artificial intelligence, including: acquiring user data corresponding to a target user in at least one characteristic domain; determining the intra-domain features corresponding to each feature domain according to the user data contained in each feature domain and a plurality of attribute vectors used for representing attribute directions; fusing the intra-domain features to obtain inter-domain features corresponding to the target users, and determining the target features according to the inter-domain features; and identifying the target characteristics to generate a user label of the target user.

According to an aspect of the present disclosure, there is provided an artificial intelligence based user tag determination apparatus, comprising: the data acquisition module is used for acquiring user data corresponding to the target user in at least one characteristic domain; the intra-domain feature determination module is used for determining intra-domain features corresponding to each feature domain according to the user data contained in each feature domain and a plurality of attribute vectors used for representing attribute directions; the target characteristic determining module is used for fusing the intra-domain characteristics to obtain inter-domain characteristics corresponding to the target user and determining target characteristics according to the inter-domain characteristics; and the label generation module is used for identifying the target characteristics and generating a user label of the target user.

In some embodiments of the present disclosure, the intra-domain feature determination module comprises: the vectorization module is used for vectorizing the user data contained in each feature domain to obtain a discrete vector corresponding to the user data of each feature domain; and the intra-domain fusion module is used for fusing the discrete vector and the attribute vectors to acquire intra-domain features corresponding to the feature domain.

In some embodiments of the present disclosure, the target feature determination module comprises: and the inter-domain fusion module is used for fusing the intra-domain features of the plurality of feature domains with the plurality of attribute vectors to obtain the inter-domain features corresponding to the target user.

In some embodiments of the present disclosure, the inter-domain feature comprises a plurality of sub-features, and each of the sub-features is mapped to a hidden-layer vector; the target feature determination module includes: the cross processing module is used for fusing the inter-domain features to obtain cross features according to two sub-features in the inter-domain features and hidden layer vectors mapped by the two sub-features; and the fusion processing module is used for carrying out feature processing on the cross features through full connection to obtain target features of the user data for representing the target user.

In some embodiments of the present disclosure, the tag generation module comprises: and the label determining module is used for identifying the target characteristics through a trained deep learning model for identifying the category to which the user data belongs, and generating the user label of the user data.

In some embodiments of the present disclosure, the tag determination module is configured to: and determining the probability of the user data belonging to each candidate label according to the target characteristics, and determining the user label according to the probability.

In some embodiments of the present disclosure, the apparatus further comprises: and the information recommendation module is used for screening a plurality of pieces of information to be selected according to the user tags to obtain target information which accords with the user tags, and pushing the target information to the terminal equipment of the target user for display.

In some embodiments of the present disclosure, the apparatus further comprises: the model training module is used for training a deep learning model according to the feature vectors of the historical data of the reference user of the at least one feature domain and the labels of the historical data to obtain the trained deep learning model; wherein the feature vector is determined according to a discrete vector of the history data and a plurality of attribute vectors for representing attribute directions of the history data.

In some embodiments of the present disclosure, the model training module is configured to: vectorizing the historical data of each feature domain to obtain a discrete vector corresponding to the historical data of each feature domain; acquiring intra-domain features of historical data according to discrete vectors corresponding to the historical data and a plurality of reference attribute vectors of the historical data; respectively combining the intra-domain features of the historical data with the plurality of reference attribute vectors to obtain inter-domain features corresponding to the plurality of reference attribute vectors; combining the inter-domain features to obtain preset features of the historical data; and training the deep learning model based on the preset features to obtain the trained deep learning model.

According to an aspect of the present disclosure, there is provided a computer readable storage medium, on which a computer program is stored, which computer program, when executed by a processor, implements the artificial intelligence based user tag determination method of any one of the above.

In the technical solutions provided by some embodiments of the present disclosure, user data corresponding to a target user is first acquired from at least one feature domain; determining the intra-domain features and the inter-domain features corresponding to each feature domain according to the user data contained in each feature domain and the attribute vectors used for representing the attribute directions so as to obtain target features; further, the target features are identified, and a user label of the target user is generated. The technical scheme of the disclosure includes that on one hand, the intra-domain features of each feature domain are obtained through the user data of at least one feature domain and the attribute vectors for representing the attribute directions, and then the inter-domain features are obtained through fusing the intra-domain features and the attribute vectors, the feature fusion is carried out on the attribute vectors and the user data when the intra-domain features of each feature domain are extracted, the inter-domain features are further determined according to the intra-domain features and are combined with the attribute vectors for representing the attribute directions to carry out the feature fusion again, the features are fused while the feature extraction is carried out, the problem that all the features are fused on the upper layer of a model after all the feature extraction is finished in the related technology is solved, the features in the user data can be comprehensively and pertinently extracted according to the attribute directions from the attribute directions specified in the fusion attribute vectors of the bottommost layer processed by the model in the feature extraction process, the features in multiple directions can be obtained according to multiple attribute vectors, information loss caused when fusion is carried out on the upper layer of the model is avoided, the problems that data in a certain direction is less and misoperation and the like caused by the fact that the direction cannot be determined are also avoided, the sufficiency and the comprehensiveness of the data can be improved, the extracted features are more comprehensive and more accurate, and the process of determining the features without damage is achieved. On the other hand, the intra-domain features and the inter-domain features of each feature domain can be acquired according to the attribute vectors and the user data, so that comprehensive and lossless target features can be obtained, user tags which accord with target users can be generated by identifying the target features, the problem of inaccurate tags caused by information loss is solved, the comprehensiveness and pertinence of the user tags are improved, the interest information of the target users can be accurately described through the accurate user tags, and the tags which are used for expressing the interest information of the target users can be generated in a lossless manner.

It is to be understood that both the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the disclosure.

Drawings

The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present disclosure and together with the description, serve to explain the principles of the disclosure. It is to be understood that the drawings in the following description are merely exemplary of the disclosure, and that other drawings may be derived from those drawings by one of ordinary skill in the art without the exercise of inventive faculty. In the drawings:

fig. 1 shows a schematic diagram of an exemplary system architecture to which technical aspects of embodiments of the present disclosure may be applied;

FIG. 2 is a schematic flow chart diagram illustrating a method for artificial intelligence based user tag determination in an embodiment of the present disclosure;

FIG. 3 schematically shows a structural schematic of a model according to one embodiment of the present disclosure;

FIG. 4 schematically shows a flow diagram of model training according to one embodiment of the present disclosure;

FIG. 5 schematically illustrates a schematic diagram of feature fusion according to one embodiment of the present disclosure;

FIG. 6 schematically shows a schematic diagram of test results according to one embodiment of the present disclosure;

FIG. 7 schematically illustrates a block diagram of an artificial intelligence based user tag determination apparatus in an embodiment of the present disclosure;

FIG. 8 illustrates a schematic structural diagram of a computer system suitable for use in implementing the electronic device of an embodiment of the present disclosure.

Detailed Description

Example embodiments will now be described more fully with reference to the accompanying drawings. Example embodiments may, however, be embodied in many different forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this disclosure will be thorough and complete, and will fully convey the concept of example embodiments to those skilled in the art.

Furthermore, the described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments. In the following description, numerous specific details are provided to give a thorough understanding of embodiments of the disclosure. One skilled in the relevant art will recognize, however, that the subject matter of the present disclosure can be practiced without one or more of the specific details, or with other methods, components, devices, steps, and so forth. In other instances, well-known methods, devices, implementations, or operations have not been shown or described in detail to avoid obscuring aspects of the disclosure.

The block diagrams shown in the figures are functional entities only and do not necessarily correspond to physically separate entities. I.e. these functional entities may be implemented in the form of software, or in one or more hardware modules or integrated circuits, or in different networks and/or processor means and/or microcontroller means.

The flow charts shown in the drawings are merely illustrative and do not necessarily include all of the contents and operations/steps, nor do they necessarily have to be performed in the order described. For example, some operations/steps may be decomposed, and some operations/steps may be combined or partially combined, so that the actual execution sequence may be changed according to the actual situation.

Fig. 1 shows a schematic diagram of an exemplary system architecture to which the technical solutions of the embodiments of the present disclosure may be applied.

As shown in fig. 1, the system architecture 100 may include a first end 101, a network 102, and a second end 103. The first end 101 may be a client, and the client may specifically be a terminal device with a display screen, such as a portable computer, a desktop computer, a smart phone, and a smart television, and is configured to display information, such as text or video, that can be viewed by a user by installing an application program or logging in a website. The number of clients may be one or more, and multiple clients may perform the same function. Network 102 serves as a medium for providing a communication link between first end 101 and second end 103. The network 102 may include various connection types, such as wired communication links, wireless communication links, etc., and in the disclosed embodiment, the network 102 between the first end 101 and the second end 103 may be a wired communication link, such as may be provided by a serial connection line, or a wireless communication link, such as may be provided by a wireless network. The second end 103 may be a client or a server as long as it has a processor disposed thereon and can perform processing operations. When the second end is a client, the second end may be the same as or different from the client of the first end. When the second end is a server, the server may be a local server or a remote server, or may be another product capable of providing a storage function or a processing function, such as a cloud server, or may be a server cluster formed by a plurality of servers, and the like, and the embodiment of the present disclosure is not particularly limited herein.

It should be understood that the number of first ends, networks and second ends in fig. 1 is merely illustrative. There may be any number of first ends, networks, and second ends, as desired for the implementation.

In an embodiment of the present disclosure, after obtaining user data according to a user behavior, the first end 101 may send the user data to the second end 103 through the network 102, and after obtaining the user data, the second end 103 may perform feature extraction on the user data to obtain a discrete vector; then, obtaining the intra-domain features and the inter-domain features of each feature domain according to the discrete vectors and the attribute vectors; then, obtaining target characteristics for identification finally through inter-domain characteristics; and finally, inputting the target feature into the model to determine the user label to which the target feature belongs. Further, in the embodiment of the present disclosure, information recommendation may also be performed on the target user after the user tag is acquired. According to the technical scheme of the embodiment of the invention, the feature extraction can be carried out on the bottom layer of the model by combining the attribute vectors, so that the accuracy can be improved, and the user experience is further improved.

It should be noted that, the method for determining a user tag based on artificial intelligence provided by the embodiment of the present disclosure may be completely executed by the second end 103 (server), may also be completely executed by the first end 101 (client), may also be partially executed by the first end, and partially executed by the second end, where an execution subject of the method for determining a user tag based on artificial intelligence is not particularly limited. Accordingly, artificial intelligence based user tag determination means may be provided in the second end 103 or in the first end 101.

In the related art, when multiple interests of a user are processed, the user often learns each feature domain independently, and information of each feature domain is directly connected at an upper layer to further mine the interests of the user. Extracting interest in the feature domain: it may become possible to extract interest within the domain using the context as a user's target's attention. But some feature domains do not behave sufficiently, which may result in that the corresponding context vector may be biased; also the orientation of different feature fields may be completely inconsistent, which also leads to information loss at the time of fusion. Inter-domain fusion of features may use context vectors to pick valuable domain features. And the single interest direction is often single, which can determine the interest direction of the user in a lossy manner.

In view of the problems in the related art, the embodiments of the present disclosure provide a method for determining a user tag based on Artificial Intelligence, which is implemented based on machine learning, which is one of Artificial Intelligence, and Artificial Intelligence (AI), which is a theory, method, technique, and application system that simulates, extends, and expands human Intelligence, senses an environment, acquires knowledge, and uses the knowledge to obtain an optimal result using a digital computer or a machine controlled by a digital computer. In other words, artificial intelligence is a comprehensive technique of computer science that attempts to understand the essence of intelligence and produce a new intelligent machine that can react in a manner similar to human intelligence. Artificial intelligence is the research of the design principle and the realization method of various intelligent machines, so that the machines have the functions of perception, reasoning and decision making.

The artificial intelligence technology is a comprehensive subject and relates to the field of extensive technology, namely the technology of a hardware level and the technology of a software level. The artificial intelligence infrastructure generally includes technologies such as sensors, dedicated artificial intelligence chips, cloud computing, distributed storage, big data processing technologies, operation/interaction systems, mechatronics, and the like. The artificial intelligence software technology mainly comprises a computer vision technology, a voice processing technology, a natural language processing technology, machine learning/deep learning and the like.

Computer Vision technology (CV) Computer Vision is a science for researching how to make a machine "see", and further refers to that a camera and a Computer are used to replace human eyes to perform machine Vision such as identification, tracking and measurement on a target, and further image processing is performed, so that the Computer processing becomes an image more suitable for human eyes to observe or transmitted to an instrument to detect. As a scientific discipline, computer vision research-related theories and techniques attempt to build artificial intelligence systems that can capture information from images or multidimensional data. Computer vision technologies generally include image processing, image recognition, image semantic understanding, image retrieval, OCR, video processing, video semantic understanding, video content/behavior recognition, three-dimensional object reconstruction, 3D technologies, virtual reality, augmented reality, synchronous positioning, map construction, and other technologies, and also include common biometric technologies such as face recognition and fingerprint recognition.

Machine Learning (ML) is a multi-domain cross discipline, and relates to a plurality of disciplines such as probability theory, statistics, approximation theory, convex analysis, algorithm complexity theory and the like. The special research on how a computer simulates or realizes the learning behavior of human beings so as to acquire new knowledge or skills and reorganize the existing knowledge structure to continuously improve the performance of the computer. Machine learning is the core of artificial intelligence, is the fundamental approach for computers to have intelligence, and is applied to all fields of artificial intelligence. Machine learning and deep learning generally include techniques such as artificial neural networks, belief networks, reinforcement learning, transfer learning, inductive learning, and formal education learning.

With the research and progress of artificial intelligence technology, the artificial intelligence technology is developed and applied in a plurality of fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, unmanned aerial vehicles, robots, smart medical care, smart customer service, and the like.

The scheme provided by the embodiment of the disclosure relates to a user tag determination technology, and is specifically explained by the following embodiment.

Fig. 2 schematically illustrates a flowchart of an artificial intelligence based user tag determination method according to an embodiment of the present disclosure, which may be performed by a server, which may be the server 103 shown in fig. 1. Referring to fig. 2, the method for determining a user tag based on artificial intelligence at least includes steps S210 to S240, which are described in detail as follows:

in step S210, user data corresponding to the target user in at least one feature field is acquired.

In one embodiment of the present disclosure, a feature field refers to a field to which user data of different attributes belongs. Such as the age domain, gender domain, video domain, or graphics domain, etc. In the embodiment of the present application, in one feature domain, the number of the user data corresponding to the target user may be one or multiple. For example, in the gender domain, the number of discrete user data corresponding to the target user identification is one, i.e., either male or female. In the video domain, the number of user data corresponding to the target user may be more than one, for example, user a watches tv show 1 and user a watches movie 2.

The target user refers to a user who generates browsing behavior by viewing information through the terminal device, and can be uniquely represented by a target user identifier. The target user identification may be an account number used by the target user to log into a certain website or application program, and the like.

The user data is discrete data generated by user behavior such as browsing behavior of the user. Since each user will typically only act on content that is of interest to him. The behavior generated based on these interesting content is then the user data for each user.

In one embodiment of the present disclosure, the division of the feature domain may be customized according to actual needs. The more than one feature domain obtained by dividing in one feature domain dividing mode may be one feature domain obtained by dividing in another feature domain dividing mode. For example, the age field, the gender field, and the region field are obtained by division in the a mode, and the basic information field is obtained by division in the B mode. Then, it can be considered that the age field, the gender field, and the region field may be sub-fields of the basic information field. One feature domain obtained by dividing in one feature domain dividing mode can also be more than one feature domain obtained by dividing in another feature domain dividing mode. For example, the video domain is obtained by division in the mode a, and the video label domain and the video category domain are obtained by division in the mode B. Then, it can be considered that the video tag field and the video category field may be sub-fields of the video field. For example, video categories in the field of video categories may be divided into sports, food, entertainment, and automobiles, etc. The video category can also be subdivided to obtain video tags, for example, sports can be subdivided into ball games, track and field games, racing games and swimming games, and the ball games can be further divided into basketball games, football games, table tennis games and the like. In one embodiment of the present disclosure, the at least one feature field may be an age field, a gender field, a geographic field, a teletext tag field, a video tag field, or the like.

In step S220, an intra-domain feature corresponding to each feature domain is determined according to the user data included in each feature domain and an attribute vector used for representing an attribute direction.

In an embodiment of the present disclosure, the attribute vector refers to a vector corresponding to a target that needs to be obtained in a specific scenario, where the specific scenario may be a search scenario, and the target may be a search target or a query target, and the target of a search task in different search scenarios may be different. The attribute vector may be represented as a query vector or a shared context, and is used to characterize an attribute direction corresponding to a search target in a certain specific scene, where the attribute direction may be understood as an interest direction, and the attribute vector in the embodiment of the present disclosure may include multiple attribute directions. An attribute vector represents an interest direction, and for a plurality of feature domains, if the attribute vector is shared among the plurality of feature domains, the interest directions of the feature domains are consistent. If there are multiple attribute vectors shared between multiple feature domains, then there may be multiple different directions of interest. For example, if there is attribute vector 1 representing sports and attribute vector 2 representing literature, then it can be considered that there are multiple directions of interest, and multiple directions of interest include sports and literature. It should be noted that the attribute vector is automatically generated in the process of model training, and does not need to be set manually. By the aid of the attribute vectors, for user data of a plurality of feature domains with large differences, features can be clearly extracted through a plurality of dimensions or a plurality of interest directions, information loss caused when the features are extracted through only one integral direction is avoided, integrity and comprehensiveness of the data and the extracted features are guaranteed, and user labels are accurately determined.

When user data is acquired, in order to improve processing efficiency, the user data of a plurality of feature domains can be processed through a trained deep learning model. Based on this, in order to improve the accuracy, the deep learning model may be trained first to obtain a well-trained deep learning model. Specifically, a deep learning model can be trained according to a feature vector of historical data of a reference user of at least one feature domain and a label of the historical data, so as to obtain the trained deep learning model; wherein the feature vector is determined according to a discrete vector of the history data and an attribute vector for representing the direction of the attribute. The reference user may include other users than the target user who generate browsing behavior by viewing information through the terminal device, and the label of the reference user can be determined according to the deep learning model and manually. The historical data may be data generated by referring to browsing behaviors of the user, the historical data may also include data of a plurality of domains, and the characteristic domain and the specific numerical value of the historical data may be the same as or different from the characteristic domain or the specific data corresponding to the target data. The label of the history data here refers to a manually labeled label, i.e., an actual label. The feature vector refers to a vector of features extracted from the historical data that can be directly used to determine the historical data and characterize the historical data. The discrete vector of the history data refers to a vector in which the history data is directly subjected to vectorization processing, but is not subjected to other processing. The attribute vector representing the attribute direction of the history data refers to a vector for representing the direction of interest of the search target determined from the history data, where the attribute vector representing the attribute direction of the history data may be determined in real time according to an actual scene, the attribute direction may be multiple, and the attribute direction of the history data may be the same as or different from the direction of interest of the user data of the target user, and is not limited herein.

The deep learning model may be any suitable model that can be used for classification, such as a convolutional neural network model, and so forth. In the embodiment of the present disclosure, a deep learning model is taken as an example of the attention mechanism. The attention mechanism may allow a neural network to focus on only a portion of its input, which may be able to select a particular input. Note that the force mechanism can be applied to any type of input, regardless of its shape, for inputs in the form of a matrix, such as an image or a vector, etc.

A network architecture diagram of a deep learning model is schematically shown in fig. 3, and referring to the diagram shown in fig. 3, the model comprises, from bottom to top: the system comprises a feature input layer, an attention fusion layer, a feature cross layer, a full connection layer and a layering layer. The characteristic input layer is used for processing data into vectors; the attention fusion layer comprises an intra-domain fusion layer and an inter-domain fusion layer and is used for realizing intra-domain fusion and inter-domain fusion; the feature crossing layer is used for fusing the features again; the full connection layer is used for fusing the features into high-order features; the layering is used to derive probabilities of predictive labels to train the model. The input end refers to historical data of a reference user, and the output end refers to the probability of which candidate label the predicted user label belongs to.

Based on the network structure diagram of the deep learning model shown in fig. 3, fig. 4 schematically shows a flowchart of a model training method in the embodiment of the present disclosure, and referring to fig. 4, the method mainly includes the following steps S410 to S450, where:

in step S410, performing vectorization processing on the historical data of each feature domain to obtain a discrete vector corresponding to the historical data of each feature domain;

in step S420, obtaining intra-domain features of the historical data according to the discrete vectors corresponding to the historical data and the plurality of reference attribute vectors;

in step S430, the intra-domain features of the historical data are respectively combined with the multiple reference attribute vectors to obtain inter-domain features corresponding to the multiple reference attribute vectors;

in step S440, combining inter-domain features of the historical data to obtain preset features of the historical data;

in step S450, the deep learning model is trained based on the preset features to obtain the trained deep learning model.

In one embodiment of the present disclosure, first, the historical data of each feature domain is input into a deep learning model; and vectorizing each historical data through an input layer of the deep learning model to obtain a discrete vector corresponding to each historical data. Specifically, after the historical data respectively corresponding to the target user identifier under the at least one feature domain is obtained, the historical data in the text form may be directly input into the deep learning model. And the input layer of the deep learning model maps each historical data into a corresponding discrete vector. For example, the input layer of the deep learning model maps the historical data "beijing city sunny zone" as "[ 102030040. ]", and the like. It should be noted that the input layer maps the history data into discrete vectors, and each feature of each feature domain is mapped into a vector of length E. Since some feature fields contain features on the order of millions, and some only tens of features, E is taken as large as possible, e.g., 128 or 256, in order to hold enough feature information.

And secondly, fusing the discrete vectors under each characteristic domain through a deep learning model attention fusion layer to obtain the intra-domain characteristics corresponding to each characteristic domain and corresponding to the target user identification. The intra-domain features may specifically be represented by intra-domain feature vectors. The method comprises the following specific steps: in an intra-domain fusion layer of the deep learning model, respectively acquiring attention distribution weights corresponding to discrete vectors under each characteristic domain; and performing linear fusion on the discrete vectors under each characteristic domain according to the attention distribution weights corresponding to the discrete vectors through the intra-domain fusion layer to obtain intra-domain characteristic vectors corresponding to the characteristic domains and corresponding to the target user identification. The intra-domain fusion layer is a network layer for fusing discrete vectors corresponding to historical data in the domain according to the feature domain in the deep learning model of the embodiment of the application. Through the intra-domain fusion layer, the discrete vectors which are input into the intra-domain fusion layer and belong to the same characteristic domain can be fused into one intra-domain characteristic, and at least one intra-domain characteristic is obtained. The number of features in the domain is the same as the number of feature domains, i.e., each feature domain corresponds to a feature in the domain. For example, historical data within a feature domain may vary in importance to the user's profile. A user has browsing records of tens of thousands of sports videos in a certain field, but only has browsing records of a few entertainment videos. When the discrete vectors corresponding to the historical data in one feature domain are fused, different attention distribution weights are distributed to different discrete vectors through an attention mechanism, and the importance degree of the discrete vectors is embodied through the attention distribution weights. The attention mechanism assigned weights can be derived from equation (1):

wherein, α_iA weight is assigned to the attention of the user,

is an attribute vector, which may also be referred to as intra-domain attention vector or inter-domain attention vector;

is an input discrete vector, W_tIs a matrix of the spatial transformation which is,

is the offset vector and H is the number of discrete vectors in the feature domain. Based on the above equation (1), each discrete vector will pass through the spatial variation matrix W_tAnd an offset vector

And transformation of the non-linear function relu into attention space, and attribute vectors

Multiplying, then performing weight calculation through the softmax layer, finally performing weighted averaging, and outputting domain fusion results of each feature domain, namely, the intra-domain features, which can be specifically shown as formula (2):

specifically, referring to fig. 3, the discrete vectors output from the input layer of the deep learning model are input into the intra-domain fusion layer, and the intra-domain fusion layer fuses the discrete vectors in each feature domain based on the attention mechanism by multiplying the discrete vectors and the attribute vectors, so as to obtain the intra-domain features corresponding to each feature domain and output the intra-domain features to the next layer. Specifically, the intra-domain fusion layer of the deep learning model can calculate an attention assignment weight for each discrete vector based on an attention mechanism through the model parameters of the deep learning model. When the attention distribution weight is determined, the weight can be distributed by combining with a reference attribute vector for representing the interest direction of a search target, so that historical data in a feature domain can be respectively attributed to different interest directions. And the intra-domain fusion layer of the deep learning model performs weighting and averaging on the discrete vectors and the weights in the characteristic domain to obtain the intra-domain characteristics corresponding to the characteristic domain.

In the formula (1), if the discrete vectors are replaced by intra-domain vectors, intra-domain features corresponding to a plurality of feature domains can be fused according to different attribute vectors by the same method, so as to obtain inter-domain features corresponding to different attribute vectors. Specifically, in an inter-domain fusion layer of the deep learning model, attention distribution weights corresponding to features in each domain are obtained by combining a plurality of attribute vectors respectively; and linearly fusing the intra-domain features according to the attention distribution weight through the inter-domain fusion layer to obtain the inter-domain features corresponding to the target user identification. The inter-domain fusion layer is a network layer for fusing the intra-domain features of each feature domain among domains in the deep learning model of the embodiment of the disclosure. The intra-domain features of the feature domains input by the intra-domain fusion layer can be fused into one inter-domain feature. For example, a user has tens of thousands of browsing records in the field of graphics, but only a few browsing records in the field of video. Then the user's interest in the teletext domain is significantly higher than in the video domain. In this embodiment, different attention distribution weights are assigned to the features in different domains through an attention mechanism, and the importance degree of the feature domain is represented through the attention distribution weights, so that the feature domain which represents the importance to the user characteristics is highlighted.

In the embodiment of the disclosure, when the discrete vectors in the feature domain are fused, different fusion weights are allocated to different discrete vectors by adopting an attention mechanism fusion mode and combining the determined multiple attribute vectors, so that more important information can be selectively selected from a large number of discrete vectors as assistance, user characteristics can be more fully and comprehensively represented, and the accuracy and the effectiveness of the obtained features in the domain are greatly improved. Similarly, when the intra-domain features corresponding to the feature domains are fused, different fusion weights are distributed to the different intra-domain features by adopting an attention mechanism fusion mode and combining the determined multiple reference attribute vectors, more important information can be selectively selected from a large number of intra-domain features as assistance, the problem of information loss is avoided, the user characteristics are more fully expressed, and the accuracy and the effectiveness of the obtained inter-domain features are improved.

Further, through a feature cross layer of the deep learning model, sparse removal processing is carried out on inter-domain features to obtain densified inter-domain features; performing second-order cross processing on the sub-features in the densified inter-domain features to obtain cross features; and deriving a preset feature for directly determining the tag based on the cross feature.

In an embodiment of the present disclosure, through a feature intersection layer of a deep learning model, performing sparseness removal on inter-domain features of historical data to obtain densified inter-domain features, including: and respectively mapping the sub-features in the inter-domain features into hidden layer vectors with preset dimensionality through a feature cross layer of the deep learning model so as to express a plurality of vectors from a linear angle. Performing second-order cross processing on sub-features in the densified inter-domain features to obtain cross features, wherein the cross features comprise: for any two sub-features in the inter-domain features, taking the product of the two sub-features and hidden layer vectors mapped by the two sub-features as second-order cross features of the two sub-features; and combining the second-order cross feature vectors to obtain cross features.

For example, the feature intersection layer of the deep learning model may combine sub-features x_iMapping to hidden vector v_iAs an expression, the sub-feature x_jMapping to hidden vector v_jAs an expression, and then (v) is calculated_i·v_j)*x_i*x_jLearning the sub-feature x_iAnd v_jThe second-order cross feature between the two can be specifically expressed as shown in formula (3):

the feature cross layer of the deep learning model can map sub-features in the inter-domain features into a hidden layer vector respectively through model parameters of the feature cross layer, so that for each sub-feature in the inter-domain features, a product of the feature cross layer and the hidden layer vector mapped by the feature cross layer can be obtained, the feature cross layer of the deep learning model performs point multiplication operation on any two products to obtain a plurality of cross sub-features, and therefore the feature cross layer of the deep learning model can splice the cross sub-features to obtain the cross features. In the embodiment of the disclosure, the intra-domain fusion layer and the inter-domain fusion layer of the deep learning model perform the first-order fusion operation, and the feature crossing layer of the deep learning model performs the second-order fusion operation. By carrying out cross processing on the sub-features in the inter-domain features, the problem that the cross features are difficult to design manually due to manual work can be avoided, so that even in a scene with complex feature distribution, the features can be easily fitted to obtain high-order cross features, the expression of the correlation of the user characteristics between different feature domains can be realized through second-order and higher-order cross features, and the accuracy of the feature vectors for finally representing the user characteristics is improved.

Further, for the high-order user features after the intersection, two layers of full connection are performed for feature fusion, and the final user high-order features, namely the preset features, are generated in the last layer. Specifically, through a first full-connection layer of the deep learning model, after the dimension of the cross feature is adjusted to be consistent with the dimension of the inter-domain feature, the cross feature is spliced with the inter-domain feature to obtain an intermediate feature. Since the inter-domain features reflect the features of the user in and between domains, and the cross features reflect the correlation among the sub-features of the inter-domain features of the user, which are all the features extracted by the preamble layer and capable of embodying the user characteristics, the first full connection layer can synthesize the features to cooperatively represent the user characteristics. Then mapping the intermediate features into preset features through a second full-connection layer of the deep learning model; and determining the label to which the historical data belongs according to the preset characteristics.

And finally, training the multi-label classifier according to the final high-order characteristics (preset characteristics) of the user by the grading layer of the deep learning model. Each user will correspond to K training samples. The sigmoid cross entropy is then used to define the loss function, which is then optimized using adam. The loss function is shown in equation (4):

wherein y is_kIs the estimated value (0-1) of the model for the kth sample,

is the true value (0 or 1) of the tag.

Through the steps in fig. 4, based on the historical data and the actual labels of the historical data, the weight parameters in the deep learning can be adjusted according to the comparison result of the deep learning model degree on the predicted labels of the historical data until the loss function approaches 0 or the deep learning model converges, so as to complete the training process of the deep learning model and obtain the trained deep learning model.

In one embodiment of the disclosure, in the process of extracting the features of the historical data of the reference user, a plurality of reference attribute vectors representing a plurality of interest directions of the target task are fused to extract the intra-domain vector of each feature domain, and then a plurality of reference attribute vectors are still fused to obtain the inter-domain features when each intra-domain vector is processed, the intra-domain features and the inter-domain features which conform to each interest direction can be respectively extracted from the angles of the plurality of interest directions, so that the problem of over-fitting or feature loss caused by insufficient data of a certain feature domain is avoided, and by sharing the context vectors and the feature inter-domain fusion vectors of the feature domains, the interest of the user can be represented in a lossless manner in the interest fusion step, meanwhile, the situation that the information of some feature domains is insufficient can be compatible, and the information of each feature domain of the user can be better mined, the comprehensiveness and the accuracy are improved, meanwhile, the accuracy of model training is improved based on the reference attribute vector, and a more accurately trained deep learning model is obtained.

In step S220, based on the trained deep learning model, the in-domain features of the user data of each feature domain may be determined by the in-domain fusion layer in the trained deep learning model. Specifically, the user data of each feature domain of the target user is input into an input layer of the deep learning model, and the user data is vectorized to obtain a discrete vector corresponding to each user data. Then inputting the discrete vectors into an intra-domain fusion layer of the deep learning model, and respectively obtaining attention distribution weights corresponding to the discrete vectors in each feature domain; and performing linear fusion on the discrete vectors under each feature domain according to the attention distribution weights corresponding to the discrete vectors through the intra-domain fusion layer to obtain intra-domain features corresponding to the feature domains and corresponding to the target user identification. When the attention distribution weight is determined, the weight can be distributed by combining with the attribute vector for representing the interest direction of the search target, so that the user data in one feature domain can be respectively classified into different interest directions. The attribute vector here may be the same as or different from the reference attribute vector, but both are used to identify the direction of interest of the search target, and the attribute vector is dynamically updated. Due to the fact that the learning of the feature domain is not sufficient, the attribute vectors in all the feature domains can be shared, the problem that some feature domains are not sufficient in behavior can be solved, all the domains face more robust training data, and the learning is more sufficient. The directions of different characteristic domains can be unified, and the fusion process is helped to express the interest of the user without damage.

Continuing to refer to fig. 2, in step S230, the intra-domain features are fused to obtain inter-domain features corresponding to the target user, and the target features are determined according to the inter-domain features.

In one embodiment of the present disclosure, an inter-domain feature refers to a feature obtained by combining intra-domain features of different feature domains. Specifically, intra-domain features of a plurality of feature domains can be respectively input into an inter-domain fusion layer of a trained deep learning model, and attention distribution weights corresponding to the intra-domain features are obtained by combining a plurality of attribute vectors of a search target of user data; and linearly fusing the intra-domain features according to the attention distribution weight through the inter-domain fusion layer to obtain the inter-domain features corresponding to the target user identification. When the intra-domain features corresponding to each feature domain are fused, different fusion weights are distributed to the different intra-domain features by adopting an attention mechanism fusion mode and combining the determined multiple attribute vectors, more important information can be selectively selected from a large number of intra-domain features as assistance, the problem of information loss is avoided, the user characteristics are more fully expressed, and the accuracy and the effectiveness of the obtained inter-domain features are improved. In the embodiment of the disclosure, the attribute vectors are combined during the bottom layer extraction and fusion, so that information loss can be avoided, and the process of determining the features without damage is realized.

After the inter-domain features are obtained, the inter-domain features can be input into a feature cross layer of the trained deep learning model. Since the sub-features of each inter-domain feature can be mapped to an implicit vector to represent a large amount of data from a linear perspective. Thus, determining a target feature from the inter-domain features comprises: fusing the inter-domain features according to the two sub-features in the inter-domain features and the hidden layer vectors mapped by the two sub-features to obtain cross features; and carrying out feature processing on the cross features through full connection to obtain target features of user data for representing target users. That is, for any two sub-features in the inter-domain features, the product of the two sub-features and the hidden layer vector mapped by the two sub-features is used as the second-order cross feature of the two sub-features; and combining the second-order cross feature vectors to obtain cross features. After the cross features are obtained, the cross features can be input into the first full connection layer, the dimension of the cross features is adjusted to be consistent with that of the inter-domain features, and then the cross features are spliced with the inter-domain features to obtain the intermediate features. Further, the intermediate features may be input into the second fully-connected layer, and the intermediate features are mapped to target features corresponding to the user data, that is, final user high-order features are generated in the last layer. By combining the attribute vector, the characteristics corresponding to the user data can be completely extracted.

A schematic diagram of extracting intra-domain features and inter-domain features is schematically shown in fig. 5. Referring to fig. 5A in fig. 5, for the tag feature domain, there are a plurality of user data, and combining discrete vectors of the user data with a shared context vector (attribute vector) enables extracting features from the user data of the tag feature domain in the same interest direction as the attribute vector based on the accurate interest direction to determine the in-domain features of the tag feature domain. Similarly, for the category feature domain of the category, a plurality of user data exist, and the discrete vectors of the user data are combined with the shared context vector (attribute vector), so that the user data of the tag feature domain can be characterized according to the same interest direction as the attribute vector to determine the intra-domain features of the category feature domain. Each feature domain may generate an intra-domain feature for each direction of interest. The plurality of attribute vectors corresponding to different feature domains can be the same, so that the plurality of feature domains have the same interest direction, and feature extraction is facilitated.

As shown by inter-domain feature extraction shown in fig. 5B in fig. 5, the label feature domain has fusion features 1 and 2, and the category feature domain also has fusion features 1 and 2, which are intra-domain features. For the label fusion feature 1 and the category fusion feature 1, the inter-domain feature between the label fusion feature 1 and the category fusion feature 1 can be obtained by combining the attribute vector 1; for the label fusion feature 2 and the category fusion feature 2, the attribute vector 2 can be combined to obtain the inter-domain features between the two. Attribute vector 1 and attribute vector 2 represent different directions of interest. In this way, a plurality of inter-domain features can be obtained based on a plurality of different interest directions. Through the technical solutions shown in fig. 3 and 5, the same attribute vector is shared between the intra-domain and the inter-domain, different attribute vectors represent different interest directions, and the images represent double-headed interest directions, so that the user interest information is extracted without loss. Especially for the interest directions with conflicts, the features corresponding to each interest direction can be more accurately and comprehensively extracted, and the lossless extraction of the features is realized.

Fig. 6 is a schematic diagram illustrating an extraction result, and referring to fig. 6, it may be determined that different interest directions represent different interest points, and when an interest is mined in a feature domain, different pieces of mining information with different pieces of mining information can be obtained, for example, if two pieces of mining information are mixed together, a part of information amount may be lost by integration of two interest directions.

Continuing to refer to fig. 2, in step S240, the target feature is subjected to recognition processing, and a user tag of the target user is generated.

In one embodiment of the present disclosure, the target feature may be input to the scoring layer of the trained deep learning model to obtain the probability that the target feature belongs to each candidate label. The trained deep learning model is used for identifying the category to which the user data belongs. The user tag may be used to depict a user representation. The user tags are used for representing categories of content interested by the user, specifically, the user tags can be identifiers corresponding to interest directions of each type, the user tags can be represented by character identifiers or identifiers of other types, and the user tags corresponding to the interest directions of different categories are different. For example, tag A corresponds to category 1, tag B corresponds to category 2, and so on.

Through the extracted target features, a user tag corresponding to a target user can be obtained, and a plurality of user tags of one target user may be included, which is not limited herein. Specifically, the probability that the user data belongs to each candidate tag is determined according to the target characteristics, and the user tag is determined according to the probability. Specifically, a probability threshold value may be set in advance; when the probability is greater than or equal to the probability threshold, it may be determined that the user data belongs to the category. Further, one or more user tags as target users may be determined from a plurality of candidate tags meeting a probability threshold in an order of descending of each probability. For example, the probability of the target user 1 for the candidate tags meets the probability threshold, and the candidate tags arranged in the order from large to small are movies, ball games, news and life in turn, and if the number of the user tags is one, the movies can be used as the user tags thereof; if the number of user tags is three, movies, ball games, news may be used as the user tags of the target user 1. In the embodiment of the disclosure, the problem that which feature domain attribute direction is one direction cannot be determined in related problems is avoided, and the direction to which each feature domain belongs can be clearly represented through the attribute vector, so that the user label of the target user can be determined without loss according to the trained deep learning model, and the accuracy is improved.

After obtaining the user tag of the target user, information may be recommended to the target user even if the target user does not generate behavior data. Specifically, a plurality of pieces of information to be selected are screened according to the user tags, target information conforming to the user tags is obtained, and the target information is pushed to terminal equipment of a target user for display. The information to be selected may be all types of information of a certain application. That is to say, the target information which meets the user label of the target user can be accurately selected from the information to be selected according to the user label of the target user, so as to be displayed, and the user experience is improved.

In the embodiment of the disclosure, the features of the user data are extracted through the trained deep learning model, and then the user label of the target user is determined according to the features, the features can be integrated to obtain the target feature which can fully reflect the characteristics of the user, and then the user label of the target user can be more accurately screened according to the target feature, so that the accuracy and efficiency of label determination are improved.

In order to determine the performance of the model, an off-line experiment is carried out on the trained deep learning model, and the off-line experiment is respectively carried out on the industry big data and the public data set so as to prove the optimized robustness, and the experiment result shows that the effectiveness of the trained deep learning model is improved relative to the basic model.

Referring to the comparison results of model performances shown in table 1, the DNN model of YOUTUBE was used as a comparison experiment, and the click log was used as a positive example and the random counter example was used as a counter example. UTPM is used as the name of a trained deep learning model, wherein AF-1head represents a model version of a single head (an interest direction); AF-2head represents a version of the model for a double head (two directions of interest). And uniformly measuring by using Prec @ K, wherein the Prec @ K represents how much proportion of the K labels with the highest scores estimated by the model is positive samples.

TABLE 1

In table 1, millions of portrait candidates are owned in the industrial dataset, so Prec @50 still has comparability, but the published dataset movielens-20M has fewer candidates, few candidates are left for selecting portrait candidates, and the feature domain has only 2 domains, but the models in two interest directions still have advantages, so that the interest mining method trained by combining the attribute vectors has robustness, the accuracy of the models is improved, and more accurate user tags can be obtained.

According to the technical scheme, by sharing the attribute vectors in the characteristic domains, the problem that some characteristic domains are not sufficient in behavior can be solved, each domain can face more robust training data, and learning is more sufficient. The interest directions of different characteristic domains can be unified, and the fusion process is helped to express the interest of the user without damage. The attribute vectors during inter-domain feature fusion are the same by multiplexing the attribute vectors learned by the intra-domain features, i.e., a plurality of attribute vectors of intra-domain fusion and inter-domain fusion are the same, so that the accuracy can be improved. Through a plurality of attribute vectors, the interest direction of which feature domain can be definitely known, the interest of the user can be expressed as losslessly as possible in the model expression process, and the accuracy is improved.

The following describes embodiments of an apparatus of the present disclosure, which may be used to perform the artificial intelligence based user tag determination method in the above embodiments of the present disclosure. For details not disclosed in the embodiments of the apparatus of the present disclosure, please refer to the embodiments of the artificial intelligence based user tag determination method described above in the present disclosure.

FIG. 7 schematically illustrates a block diagram of an artificial intelligence based user tag determination apparatus according to one embodiment of the present disclosure.

Referring to fig. 7, an artificial intelligence based user tag determination apparatus 700 according to an embodiment of the present disclosure includes: a data obtaining module 701, configured to obtain user data corresponding to a target user in at least one feature domain; an intra-domain feature determining module 702, configured to determine intra-domain features corresponding to each feature domain according to the user data included in each feature domain and a plurality of attribute vectors indicating attribute directions; a target feature determining module 703, configured to fuse the intra-domain features to obtain inter-domain features corresponding to the target user, and determine target features according to the inter-domain features; a tag generating module 704, configured to perform identification processing on the target feature, and generate a user tag of the target user.

In some embodiments of the present disclosure, based on the foregoing solution, the intra-domain feature determination module includes: the vectorization module is used for vectorizing the user data contained in each feature domain to obtain a discrete vector corresponding to the user data of each feature domain; and the intra-domain fusion module is used for fusing the discrete vector and the attribute vectors to acquire intra-domain features corresponding to the feature domain.

In some embodiments of the present disclosure, based on the foregoing, the target feature determination module includes: and the inter-domain fusion module is used for fusing the intra-domain features of the plurality of feature domains with the plurality of attribute vectors to obtain the inter-domain features corresponding to the target user.

In some embodiments of the present disclosure, based on the foregoing scheme, the inter-domain feature includes a plurality of sub-features, and each of the sub-features is mapped to a hidden layer vector; the target feature determination module includes: the cross processing module is used for fusing the inter-domain features to obtain cross features according to two sub-features in the inter-domain features and hidden layer vectors mapped by the two sub-features; and the fusion processing module is used for carrying out feature processing on the cross features through full connection to obtain target features of the user data for representing the target user.

In some embodiments of the present disclosure, based on the foregoing solution, the tag generation module includes: and the label determining module is used for identifying the target characteristics through a trained deep learning model for identifying the category to which the user data belongs, and generating the user label of the user data.

In some embodiments of the present disclosure, based on the foregoing, the tag determination module is configured to: and determining the probability of the user data belonging to each candidate label according to the target characteristics, and determining the user label according to the probability.

In some embodiments of the present disclosure, based on the foregoing solution, the apparatus further includes: and the information recommendation module is used for screening a plurality of pieces of information to be selected according to the user tags to obtain target information which accords with the user tags, and pushing the target information to the terminal equipment of the target user for display.

In some embodiments of the present disclosure, based on the foregoing solution, the apparatus further includes: the model training module is used for training a deep learning model according to the feature vectors of the historical data of the reference user of the at least one feature domain and the labels of the historical data to obtain the trained deep learning model; wherein the feature vector is determined according to a discrete vector of the history data and a plurality of attribute vectors for representing attribute directions of the history data.

In some embodiments of the present disclosure, based on the foregoing, the model training module is configured to: vectorizing the historical data of each feature domain to obtain a discrete vector corresponding to the historical data of each feature domain; acquiring intra-domain features of historical data according to discrete vectors corresponding to the historical data and a plurality of reference attribute vectors of the historical data; respectively combining the intra-domain features of the historical data with the plurality of reference attribute vectors to obtain inter-domain features corresponding to the plurality of reference attribute vectors; combining the inter-domain features to obtain preset features of the historical data; and training the deep learning model based on the preset features to obtain the trained deep learning model.

It should be noted that the computer system 800 of the electronic device shown in fig. 8 is only an example, and should not bring any limitation to the functions and the scope of the application of the embodiments of the present disclosure.

As shown in fig. 8, a computer system 800 includes a Central Processing Unit (CPU)801 that can perform various appropriate actions and processes according to a program stored in a Read-Only Memory (ROM) 802 or a program loaded from a storage section 808 into a Random Access Memory (RAM) 803. In the RAM 803, various programs and data necessary for system operation are also stored. The CPU 801, ROM802, and RAM 803 are connected to each other via a bus 804. An Input/Output (I/O) interface 805 is also connected to bus 804.

The following components are connected to the I/O interface 805: an input portion 806 including a keyboard, a mouse, and the like; an output section 807 including a Cathode Ray Tube (CRT), a Liquid Crystal Display (LCD), and a speaker; a storage section 808 including a hard disk and the like; and a communication section 809 including a Network interface card such as a LAN (Local Area Network) card, a modem, or the like. The communication section 809 performs communication processing via a network such as the internet. A drive 810 is also connected to the I/O interface 805 as necessary. A removable medium 811 such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, or the like is mounted on the drive 810 as necessary, so that a computer program read out therefrom is mounted on the storage section 808 as necessary.

In particular, the processes described below with reference to the flowcharts may be implemented as computer software programs, according to embodiments of the present disclosure. For example, embodiments of the present disclosure include a computer program product comprising a computer program embodied on a computer readable storage medium, the computer program containing program code for performing the method illustrated by the flow chart. In such an embodiment, the computer program can be downloaded and installed from a network through the communication section 809 and/or installed from the removable medium 811. When the computer program is executed by the Central Processing Unit (CPU)801, various functions defined in the system of the present application are executed. In some embodiments, the computer system 800 may further include an AI (artificial intelligence) processor for processing computing operations related to deep learning.

It should be noted that the computer readable storage medium shown in the embodiments of the present disclosure may be a computer readable signal medium or a computer readable storage medium or any combination of the two. A computer readable storage medium may be, for example, but not limited to, an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination of the foregoing. More specific examples of the computer readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer diskette, a hard disk, a Random Access Memory (RAM), a Read-Only Memory (ROM), an Erasable Programmable Read-Only Memory (EPROM), a flash Memory, an optical fiber, a portable Compact Disc Read-Only Memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing. In the present disclosure, a computer readable storage medium may be any tangible medium that can contain, or store a program for use by or in connection with an instruction execution system, apparatus, or device. In contrast, in the present disclosure, a computer-readable signal medium may include a propagated data signal with computer-readable program code embodied therein, for example, in baseband or as part of a carrier wave. Such a propagated data signal may take many forms, including, but not limited to, electro-magnetic, optical, or any suitable combination thereof. A computer readable signal medium may also be any computer readable medium that is not a computer readable storage medium and that can communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device. Program code embodied on a computer readable medium may be transmitted using any appropriate medium, including but not limited to: wireless, wired, etc., or any suitable combination of the foregoing.

The flowchart and block diagrams in the figures illustrate the architecture, functionality, and operation of possible implementations of systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each block in the flowchart or block diagrams may represent a module, segment, or portion of code, which comprises one or more executable instructions for implementing the specified logical function(s). It should also be noted that, in some alternative implementations, the functions noted in the block may occur out of the order noted in the figures. For example, two blocks shown in succession may, in fact, be executed substantially concurrently, or the blocks may sometimes be executed in the reverse order, depending upon the functionality involved. It will also be noted that each block of the block diagrams or flowchart illustration, and combinations of blocks in the block diagrams or flowchart illustration, can be implemented by special purpose hardware-based systems which perform the specified functions or acts, or combinations of special purpose hardware and computer instructions.

The units described in the embodiments of the present disclosure may be implemented by software, or may be implemented by hardware, and the described units may also be disposed in a processor. Wherein the names of the elements do not in some way constitute a limitation on the elements themselves.

As another aspect, the present application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or may exist separately without being assembled into the electronic device. The computer-readable storage medium carries one or more programs which, when executed by an electronic device, cause the electronic device to implement the method described in the above embodiments.

It should be noted that although in the above detailed description several modules or units of the device for action execution are mentioned, such a division is not mandatory. Indeed, the features and functionality of two or more modules or units described above may be embodied in one module or unit, according to embodiments of the present disclosure. Conversely, the features and functions of one module or unit described above may be further divided into embodiments by a plurality of modules or units.

Through the above description of the embodiments, those skilled in the art will readily understand that the exemplary embodiments described herein may be implemented by software, or by software in combination with necessary hardware. Therefore, the technical solution according to the embodiments of the present disclosure may be embodied in the form of a software product, which may be stored in a non-volatile storage medium (which may be a CD-ROM, a usb disk, a removable hard disk, etc.) or on a network, and includes several instructions to enable a computing device (which may be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the embodiments of the present disclosure.

Other embodiments of the disclosure will be apparent to those skilled in the art from consideration of the specification and practice of the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the disclosure following, in general, the principles of the disclosure and including such departures from the present disclosure as come within known or customary practice within the art to which the disclosure pertains.

It will be understood that the present disclosure is not limited to the precise arrangements described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present disclosure is limited only by the appended claims.

Claims

1. A user label determination method based on artificial intelligence is characterized by comprising the following steps:

acquiring user data corresponding to a target user in at least one characteristic domain;

determining the intra-domain features corresponding to each feature domain according to the user data contained in each feature domain and a plurality of attribute vectors used for representing attribute directions;

fusing the intra-domain features to obtain inter-domain features corresponding to the target users, and determining the target features according to the inter-domain features;

and identifying the target characteristics to generate a user label of the target user.

2. The method of claim 1, wherein determining the in-domain feature corresponding to each feature domain according to the user data included in each feature domain and a plurality of attribute vectors indicating attribute directions comprises:

vectorizing the user data contained in each feature domain to obtain a discrete vector corresponding to the user data of each feature domain;

and fusing the discrete vector and the attribute vectors to obtain the intra-domain features corresponding to the feature domain.

3. The method for determining the user tag based on the artificial intelligence according to claim 1, wherein the step of fusing the intra-domain features to obtain the inter-domain features corresponding to the target user comprises:

and fusing the intra-domain features of the plurality of feature domains with the plurality of attribute vectors to obtain the inter-domain features corresponding to the target user.

4. The artificial intelligence based user tag determination method of claim 1, wherein the inter-domain features comprise a plurality of sub-features, and each of the sub-features is mapped to a hidden-layer vector;

determining the target feature according to the inter-domain feature comprises:

fusing the inter-domain features to obtain cross features according to two sub-features in the inter-domain features and hidden layer vectors mapped by the two sub-features;

and carrying out feature processing on the cross features through full connection to obtain target features of the user data for representing the target user.

5. The artificial intelligence based user tag determination method according to claim 1, wherein the identifying the target feature to generate the user tag of the target user comprises:

and identifying the target characteristics through a trained deep learning model for identifying the category to which the user data belongs, and generating the user label of the user data.

6. The artificial intelligence based user tag determination method of claim 5, wherein generating the user tag of the user data comprises:

and determining the probability of the user data belonging to each candidate label according to the target characteristics, and determining the user label according to the probability.

7. The artificial intelligence based user tag determination method of claim 1, further comprising:

and screening a plurality of pieces of information to be selected according to the user tags to obtain target information which accords with the user tags, and pushing the target information to terminal equipment of the target user for display.

8. The artificial intelligence based user tag determination method of claim 5, further comprising:

training a deep learning model according to the feature vector of the historical data of the reference user of the at least one feature domain and the label of the historical data to obtain the trained deep learning model;

wherein the feature vector is determined from a discrete vector of the historical data and a plurality of reference attribute vectors representing directions of attributes of the historical data.

9. The method of claim 8, wherein training a deep learning model according to the feature vectors of the historical data of the reference user of the at least one feature domain and the tags of the historical data comprises:

vectorizing the historical data of each feature domain to obtain a discrete vector corresponding to the historical data of each feature domain;

acquiring intra-domain features of the historical data according to the discrete vectors corresponding to the historical data and the plurality of reference attribute vectors;

respectively combining the intra-domain features of the historical data with the plurality of reference attribute vectors to obtain inter-domain features corresponding to the plurality of reference attribute vectors;

combining inter-domain characteristics of the historical data to obtain preset characteristics of the historical data;

and training the deep learning model based on the preset features to obtain the trained deep learning model.

10. An artificial intelligence based user tag determination apparatus, comprising:

the data acquisition module is used for acquiring user data corresponding to the target user in at least one characteristic domain;

the intra-domain feature determination module is used for determining intra-domain features corresponding to each feature domain according to the user data contained in each feature domain and a plurality of attribute vectors used for representing attribute directions;

the target characteristic determining module is used for fusing the intra-domain characteristics to obtain inter-domain characteristics corresponding to the target user and determining target characteristics according to the inter-domain characteristics;

and the label generation module is used for identifying the target characteristics and generating a user label of the target user.

11. A computer-readable storage medium, on which a computer program is stored, which, when being executed by a processor, carries out the artificial intelligence based user tag determination method of any one of claims 1 to 9.