CN110807150B - Information processing method and device, electronic equipment and computer readable storage medium - Google Patents

Abstract

The embodiment of the invention provides an information processing method and device, electronic equipment and a computer readable storage medium, and belongs to the technical field of computers and communication. The method comprises the following steps: determining a first character of the target object; acquiring first object data of the target object according to first character attribute information of the first character; acquiring article attribute information of an object to be recommended; and processing the first object data and the item attribute information through a neural network model, and determining a first target item recommended to a first character of the target object from the to-be-recommended objects. The technical scheme of the embodiment of the invention provides an information processing method which can improve the accuracy of personalized recommendation of an object to be recommended.

Description

Information processing method and device, electronic device and computer readable storage medium

Technical Field

The present invention relates to the field of computer and communication technology, and in particular to an information processing method and device, an electronic device, and a computer-readable storage medium.

Background technique

In recent years, web games have developed rapidly and have become an important entertainment activity. Games can not only provide players with opportunities to participate, but also enhance the fun of games, improve user experience, enhance user stickiness, and increase the revenue of game platforms by selling virtual goods (such as props).

There are often many types of virtual goods in games. At the same time, games have high real-time requirements, and users need to quickly find the target props they want. Therefore, it is very important for the game platform to provide users with an efficient and accurate information screening system.

It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present invention, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field.

Summary of the invention

Embodiments of the present invention provide an information processing method and device, an electronic device, and a computer-readable storage medium, which can improve the accuracy of personalized recommendations for items to be recommended.

Other features and advantages of the present invention will become apparent from the following detailed description, or may be learned in part by practice of the present invention.

An embodiment of the present invention provides an information processing method, which includes: determining a first role of a target object; obtaining first object data of the target object based on first role attribute information of the first role; obtaining item attribute information of an object to be recommended; processing the first object data and the item attribute information through a neural network model, and determining a first target item recommended to the first role of the target object from the objects to be recommended.

An embodiment of the present invention provides an information processing device, which includes: a first object role determination module, configured to determine a first role of a target object; a first object data acquisition module, configured to obtain first object data of the target object based on first role attribute information of the first role; an item attribute information acquisition module, configured to obtain item attribute information of an item to be recommended; and a first target item determination module, configured to process the first object data and the item attribute information through a neural network model, and determine a first target item recommended to the first role of the target object from the items to be recommended.

An embodiment of the present invention provides a computer-readable storage medium on which a computer program is stored. When the program is executed by a processor, the information processing method described in the above embodiment is implemented.

An embodiment of the present invention provides an electronic device, comprising: one or more processors; a storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, the one or more processors implement the information processing method as described in the above embodiments.

In the technical solutions provided by some embodiments of the present invention, by determining the first role of the target object, and according to the first role attribute information of the first role, obtaining the first object data of the target object, and also obtaining the item attribute information of the object to be recommended, the first object data and the item attribute information can be processed by a neural network model, and the first target item recommended to the first role of the target object can be determined from the objects to be recommended. On the one hand, by comprehensively considering the rich role attribute information of the target object and the rich item attribute information of the object to be recommended to determine the target item among the objects to be recommended, the accuracy of personalized recommendation of the items can be improved; on the other hand, by using a nonlinear neural network model to characterize the complex interaction between the target object and the object to be recommended, the purpose of enhancing the model interpretability and improving the recommendation effect can be achieved.

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

BRIEF DESCRIPTION OF THE DRAWINGS

The drawings herein are incorporated into and constitute a part of the specification, showing embodiments consistent with the present invention, and together with the specification, are used to explain the principles of the present invention. Obviously, the drawings described below are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative work. In the drawings:

FIG1 is a schematic diagram showing an exemplary system architecture to which an information processing method or an information processing device according to an embodiment of the present invention can be applied;

FIG2 is a schematic diagram showing the structure of a computer system of an electronic device suitable for implementing an embodiment of the present invention;

FIG3 schematically shows a flow chart of an information processing method according to an embodiment of the present invention;

FIG4 schematically shows a flow chart of an information processing method according to another embodiment of the present invention;

FIG5 schematically shows a flow chart of an information processing method according to another embodiment of the present invention;

FIG6 is a schematic diagram showing the processing process of step S320 shown in FIG3 in one embodiment;

FIG. 7 is a schematic diagram showing the processing process of step S340 shown in FIG. 3 in one embodiment;

FIG8 is a schematic diagram showing the processing process of step S341 shown in FIG7 in one embodiment;

FIG9 schematically shows a schematic diagram of a first embedding sub-model according to an embodiment of the present invention;

FIG10 is a schematic diagram showing the processing process of step S342 shown in FIG7 in one embodiment;

FIG11 schematically shows a schematic diagram of a second embedding sub-model according to an embodiment of the present invention;

FIG12 schematically shows a structural diagram of a neural network model according to an embodiment of the present invention;

FIG13 is a schematic diagram showing the processing process of step S343 shown in FIG7 in one embodiment;

FIG14 schematically shows a schematic structural diagram of a first neural network sub-model according to an embodiment of the present invention;

FIG. 15 is a schematic diagram showing the processing process of step S344 shown in FIG. 7 in one embodiment;

FIG16 schematically shows a schematic diagram of the structure of a second neural network sub-model according to an embodiment of the present invention;

FIG. 17 is a schematic diagram showing the processing process of step S345 shown in FIG. 7 in one embodiment;

FIG18 schematically shows a schematic diagram of an interface for introducing martial arts schools according to an embodiment of the present invention;

19 and 20 schematically show a schematic diagram of an interface for customizing character details according to an embodiment of the present invention;

FIG21 schematically shows a schematic diagram of an interface after character creation is completed according to an embodiment of the present invention;

FIG22 schematically shows a schematic diagram of a game interface according to an embodiment of the present invention;

FIG23 schematically shows a schematic diagram of an interface of role attributes according to an embodiment of the present invention;

FIG24 schematically shows a schematic diagram of an interface of a game mall according to an embodiment of the present invention;

FIG. 25 schematically shows a block diagram of an information processing device according to an embodiment of the present invention.

Detailed ways

Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in many forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the present invention will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.

In addition, the described features, structures or characteristics may be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present invention. However, those skilled in the art will appreciate that the technical solution of the present invention may be practiced without one or more of the specific details, or other methods, components, devices, steps, etc. may be adopted. In other cases, known methods, devices, implementations or operations are not shown or described in detail to avoid blurring various aspects of the present invention.

The block diagrams shown in the accompanying drawings are merely functional entities and do not necessarily correspond to physically independent entities. That is, these functional entities may be implemented in software form, or in one or more hardware modules or integrated circuits, or in different networks and/or processor devices and/or microcontroller devices.

The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations/steps, nor must they be executed in the order described. For example, some operations/steps can be decomposed, and some operations/steps can be combined or partially combined, so the actual execution order may change according to actual conditions.

FIG. 1 is a schematic diagram showing an exemplary system architecture 100 to which an information processing method or an information processing device according to an embodiment of the present invention may be applied.

As shown in Fig. 1, the system architecture 100 may include one or more of terminal devices 101, 102, 103, a network 104, and a server 105. The network 104 is used to provide a medium for communication links between the terminal devices 101, 102, 103 and the server 105. The network 104 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

It should be understood that the number of terminal devices, networks and servers in FIG1 is only illustrative. According to implementation requirements, there may be any number of terminal devices, networks and servers. For example, the server 105 may be a server cluster composed of multiple servers.

Users can use terminal devices 101, 102, 103 to interact with server 105 through network 104 to receive or send messages, etc. Terminal devices 101, 102, 103 can be various electronic devices with display screens and supporting web browsing, including but not limited to smart phones, tablet computers, laptops, desktop computers, wearable devices, smart home devices, etc.

The server 105 may be a server that provides various services. For example, a user uses the terminal device 103 (or the terminal device 101 or 102) to open a game client, and determines his game character in the game on the game client, and sends a request to the server 105. The server 105 can obtain the role attribute information of the corresponding game character based on the game character information carried in the request, thereby obtaining the object data of the user; at the same time, the server 105 can also obtain the item attribute information of the items to be recommended in the game mall, process the user's object data and the item attribute information of each item to be recommended through the neural network model, determine the target item from these items to be recommended, and return the target item to the terminal device 103, so that the user can view the target item recommended to the currently selected game character on the terminal device 103.

For another example, the terminal device 103 (or the terminal device 101 or 102) may be a smart TV, a VR (Virtual Reality)/AR (Augmented Reality) helmet display, or a mobile terminal such as a smart phone, a tablet computer, etc. on which instant messaging, navigation, video applications (application, APP), etc. are installed. The user may send various requests to the server 105 through the smart TV, VR/AR helmet display, or the instant messaging, video APP. Based on the request, the server 105 may obtain feedback information in response to the request and return it to the smart TV, VR/AR helmet display, or the instant messaging, video APP, and then display the returned feedback information through the smart TV, VR/AR helmet display, or the instant messaging, video APP.

FIG. 2 shows a schematic diagram of the structure of a computer system of an electronic device suitable for implementing an embodiment of the present invention.

It should be noted that the computer system 200 of the electronic device shown in FIG. 2 is only an example and should not bring any limitation to the functions and scope of use of the embodiments of the present invention.

As shown in FIG2 , the computer system 200 includes a central processing unit (CPU) 201, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 202 or a program loaded from a storage portion 208 to a random access memory (RAM) 203. Various programs and data required for system operation are also stored in the RAM 203. The CPU 201, the ROM 202, and the RAM 203 are connected to each other via a bus 204. An input/output (I/O) interface 205 is also connected to the bus 204.

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

In particular, according to an embodiment of the present invention, the process described below with reference to the flowchart can be implemented as a computer software program. For example, an embodiment of the present invention includes a computer program product, which includes a computer program carried on a computer-readable storage medium, and the computer program contains program code for executing the method shown in the flowchart. In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 209, and/or installed from the removable medium 211. When the computer program is executed by the central processing unit (CPU) 201, various functions defined in the method and/or device of the present application are executed.

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

The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the methods, devices and computer program products according to various embodiments of the present invention. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of boxes in the block diagram or flow chart, can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

The modules and/or units and/or sub-units involved in the embodiments of the present invention may be implemented by software or hardware, and the modules and/or units and/or sub-units described may also be arranged in a processor. The names of these modules and/or units and/or sub-units do not, in certain cases, limit the modules and/or units and/or sub-units 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 embodiment; or it may exist independently without being assembled into the electronic device. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed by an electronic device, the electronic device implements the method described in the following embodiments. For example, the electronic device can implement the various steps shown in Figure 3 or Figure 4 or Figure 5 or Figure 6 or Figure 7 or Figure 8 or Figure 10 or Figure 13 or Figure 15 or Figure 17.

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.

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

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.

With the research and advancement of artificial intelligence technology, artificial intelligence technology has been studied and applied in many fields, such as common smart homes, smart wearable devices, virtual assistants, smart speakers, smart marketing, unmanned driving, automatic driving, drones, robots, smart medical care, smart customer service, etc. I believe that with the development of technology, artificial intelligence technology will be applied in more fields and play an increasingly important role.

The solution provided by the embodiment of the present invention involves artificial intelligence machine learning and other technologies, which are specifically described by the following embodiments:

First, some terms involved in the following embodiments are explained.

Deep learning: Deep learning is a learning process based on deep neural networks, which uses optimization methods such as gradient descent to ultimately obtain a one-to-one mapping function between input data and target data. For example, given a person's age, gender and other information, we hope to use this information to predict whether the person likes pets. The input data is the person's age, gender and other information, and the target data is "whether the person likes pets." What is needed is to build a correct mapping function that can map the input data to the target data. In this way, given another person's information, we can know whether the person likes pets. Deep learning is the process of solving this mapping function based on a large amount of data, and using a deep neural network to simulate this function.

Collaborative Filtering (CF): Collaborative filtering is a type of model algorithm used in recommendation systems. The core of collaborative filtering is to infer which users (or products) are more similar through past data. Therefore, this type of model can be simply divided into "user-based" and "item-based". The former looks for which users are more similar, and the latter looks for which products are more similar. For the former, after knowing similar users, the products that the user has purchased before can be recommended to other users who are similar to the user. For the latter, after knowing similar products, it can be inferred that a user likes other products similar to the product based on the fact that the user purchased the product, and then other similar products can be recommended to the user.

Metric: A custom mapping function that meets the four requirements of a metric and is used to measure the distance between any two points in the domain space. In simple terms, defining a new metric is defining a new distance.

In the embodiments of the present invention, the game scenario is used as an example for illustration, but it can be understood that the technical solution provided by the embodiments of the present invention is not limited to application in the game scenario, and can also be applied to other recommendation scenarios.

Among them, the data in the game recommendation scenario has the following characteristics:

First, the data is massive. Game platforms provide services to millions to tens of millions of users. When playing games, users’ attributes and behaviors will change dynamically, and they will choose to purchase different game products over time, which leads to the accumulation of massive user behavior data on the game platform every day.

Second, data is high-dimensional. The data in the game is high-dimensional. In addition to the user's personal portrait (such as the user's real age, gender, geographical location, and other real attribute information of the user), the attributes of the game character selected by the user also include the name, gender, occupation, and other attribute information of the game character. At the same time, there are many categories of game products, and they also have multi-dimensional attributes.

Third, complex structural associations. Compared with e-commerce and social platforms, the relationship between users and the products they purchase on game platforms is not a simple one-to-one correspondence. For example, a user can use different game characters at different times. For example, when using game character A, the user may frequently purchase game product a, but may not purchase game product a when using game character B. That is, the choices made by the same user may be different when using different game characters.

Based on the above, when designing the recommendation method in the game scenario, it is necessary to consider the data scale and data characteristics in this scenario.

Collaborative filtering is a type of recommendation algorithm in related technologies. It uses collaborative information in large amounts of data to make recommendations based on the following two starting points: (1) users with similar interests may be interested in the same things; (2) users may prefer things that are similar to what they have already purchased.

However, traditional recommendation methods based on collaborative filtering often only model the ID (identification) and rating matrix of users and products when learning vector representations of users and products, ignoring or not fully utilizing rich background information. That is, in collaborative filtering, if a user in the rating matrix purchases a certain product, the corresponding rating value is 1, otherwise it is 0, but the attribute information of each dimension of users and products is not considered.

CTR (Click-through rate prediction) is not only used for click-through rate estimation, but also has applications in scenarios such as product ranking in recommendation systems. Compared with the CF model, CTR directly calculates the probability of the target product being clicked by the target user through input data features. Its general formula can be expressed as:

y＝f(X) (1)

Where X is the input data feature matrix, and y is the probability value between [0,1], indicating the probability of the input data being clicked. Related technologies such as logistics regression (LR) and factorization machine (FM) can be used. The LR model is calculated as follows:

Among them, θ is a learnable parameter. The features of each dimension in the LR model are independent of each other, and the combination of features often helps to improve the prediction effect. Based on this, FM takes the intersection between features into account. When considering the intersection of two features, the FM model formula is as follows:

Where n represents the dimension of the input data X, n is a positive integer greater than or equal to 1, _xi and _xj represent the values of the i-th and j-th bits in X, respectively, and their values can be, for example, 0 or 1. θ and _wij are learnable parameters obtained during the model training process.

Although feature cross-pollination can improve prediction performance, a simple vector inner product operation is still used to calculate the probability of a user purchasing a product. Such traditional machine learning methods can only model the linear relationship between features, and the model capabilities have certain limitations. The predicted score value is directly calculated based on the features, and the collaborative information of users or products will not be utilized.

FIG3 schematically shows a flow chart of an information processing method according to an embodiment of the present invention. The method provided by the embodiment of the present invention can be executed by any electronic device with computing and processing capabilities, such as one or more of the terminal devices 101, 102, 103 and/or the server 105 in FIG1. In the following examples, the terminal device is used as the execution subject for example.

As shown in FIG. 3 , the information processing method provided by the embodiment of the present invention may include the following steps.

In step S310, a first role of a target object is determined.

In the embodiment of the present invention, the target object may be a game player of a game on a game platform. After the game player logs into the game platform with his game account, he may select a game character A from multiple game characters of the game as his first character at the current time t1.

In step S320, first object data of the target object is obtained according to the first role attribute information of the first role.

In an exemplary embodiment, the first role attribute information of the first role may include any one or more of identity information, current level information and/or current operation information of the first role.

Here we take a certain RPG (Role-playing game) game as an example. Assume that it has pre-set seven identities: escort, constable, hunter, killer, musician, ranger and scholar. Each identity corresponds to a unique profession in the martial arts world, and the target object can choose to join one of them.

In this RPG game, each identity can be further divided into three levels as advanced identities, and each level of advancement can gain more powerful manufacturing skills and identity skills. Players can obtain experience points by completing daily tasks, and then consume experience points to learn more relevant skills and advance their identities. For example, the identity of a scholar can include three levels in ascending order: scholar, elegant scholar, and state scholar. The scholar can be set as the lowest level and the state scholar as the highest level. The identity of a musician can include three levels in ascending order: musician, actor, and famous actor. The identity of a killer can include three levels in ascending order: killer, assassin, and killer god. The identity of a constable can include three levels in ascending order: constable, head of the constable, and god of the constable. The identity of a dart master can include three levels in ascending order: dart master, head of the dart, and divine dart. The identity of a ranger can include three levels in ascending order: ranger, chivalrous man, and heroic man. The identity of a hunter can include three levels in ascending order: hunter, arrow hunter, and hunter saint.

In this RPG game, game characters with different identities can perform different operations (such as tasks and/or gameplay), and game characters with the same identity can also perform different operations at different levels.

It should be understood that the above-mentioned identity information, current level information, current operation information, etc. are only used for illustration, and the first character attribute information can be adaptively adjusted in different application scenarios, or in other games of different types, or in other games of the same type.

In step S330, item attribute information of the item to be recommended is obtained.

Taking the game scene as an example, the recommended items can be props in the game mall of the target game selected by the target object. At this time, the item attribute information can include any one or more of the categories to which each item belongs, the functions that can be completed, the identity of the corresponding game character, the level of the corresponding game character, and the operation of the corresponding game character. For example, it can be a prop used by the game character to practice Qinggong, or a prop used by the game character to improve internal strength, etc. Specifically, the item attribute information can be set according to actual needs.

In step S340, the first object data and the item attribute information are processed by a neural network model to determine a first target item recommended to the first character of the target object from the items to be recommended.

The information processing method provided in the embodiment of the present invention determines the first role of the target object, and obtains the first object data of the target object based on the first role attribute information of the first role, and also obtains the item attribute information of the object to be recommended, so that the first object data and the item attribute information can be processed by a neural network model to determine the first target item recommended to the first role of the target object from the objects to be recommended. On the one hand, by comprehensively considering the rich role attribute information of the target object and the rich item attribute information of the objects to be recommended to determine the target items among the objects to be recommended, the accuracy of personalized recommendation of items can be improved; on the other hand, by using a nonlinear neural network model to characterize the complex interaction between the target object and the objects to be recommended, the purpose of enhancing the model interpretability and improving the recommendation effect can be achieved.

Fig. 4 schematically shows a flow chart of an information processing method according to another embodiment of the present invention. As shown in Fig. 4, compared with the above embodiment, the method provided by the embodiment of the present invention is different in that it can further include the following steps.

In step S410, a second role of the target object is determined.

For example, at another time t2, the game player may switch to another game character B as the second character.

In step S420, second object data of the target object is obtained according to the second role attribute information of the second role.

In an exemplary embodiment, the second role attribute information of the second role may include any one or more of identity information, current level information and/or current operation information of the second role.

In step S430, the second object data and the item attribute information are processed by the neural network model to determine a second target item recommended to the second character of the target object from the items to be recommended.

In the embodiment of the present invention, the same user (e.g., a game player) can select different game characters in the same game at different times. For example, if the game character A is selected at time t1, which is a low-level escort, the system will recommend the props required by the low-level escort; if the game character B is selected at time t2, which is a high-level killer, the system will recommend the props required by the high-level killer, that is, the user selects different game characters, which may have attributes such as belonging to different sects and different identities, and they will perform corresponding operations. For example, one type of props is required for practicing magic, and another type of props is required for refining the mind. A prop pool can be pre-set, and there are various props in the prop pool, each prop has corresponding item attribute information. When the game character selected by the user changes and/or the attribute value of the selected game character changes, the type of props recommended by the system to the user can change, or the arrangement order of the props recommended to the user changes. In the embodiment of the present invention, the same user may make different choices for the same game commodity under different game characters. This is because the game commodity in the game scene is the prop of the game character, and the purchase behavior is associated with the character attribute. In addition, if the attributes of the same user's game character change, the choice he makes for the same game product may also be different. For example, when the character is at a low level, he may not need to buy a luxurious item, but as the level increases, he may purchase the item.

Fig. 5 schematically shows a flow chart of an information processing method according to another embodiment of the present invention. As shown in Fig. 5, compared with the above embodiment, the method provided by the embodiment of the present invention is different in that it can further include the following steps.

In step S510, the historical role of the target object is obtained.

In step S520, historical object data of the target object is obtained according to the historical role attribute information of the historical role.

In step S530, the historical operation record of the target object on the object to be recommended is obtained.

In step S540, labels of the historical object data and the item attribute information are determined according to the historical operation records to generate a training data set.

In the embodiment of the present invention, the neural network model used can be trained in advance. When training the model, a training data set is first constructed. Here, the historical operation record of the target object on the recommended item is used. Taking the above game scene as an example, the recommended item can be each item in the item pool, and the historical operation record can be the purchase record of each item by the game player in history, but the present invention is not limited to this, and can also be the game player's historical use record, preview record, etc. of each item.

In the embodiment of the present invention, the training data set may be constructed in the following form:

The data can be divided into two parts: the first object data consisting of the target object and its first role attribute information The recommended items and their item attribute information The fields of user-related data are: [user_id, user_feature_1, user_feature_2, ...], and the fields of product-related data are: [item_id, item_feature_1, item_feature_2, ...], where the id and each feature field are represented by a numerical value, and [feature_1, feature_2, ...] is called attribute information (context). The features of users in the game are not fixed, but change over time, so the same id can correspond to multiple attribute information.

(1) It is the concatenation of the target object id and the first character attribute information of the first character of the target object. Because the same target object can be in different game roles at different times, one target object can correspond to multiple For example, when the target object id = 1, the corresponding context = [0.1, 0.3, ..., 0.2] When id＝1，context＝[0.5,0.1,…,0.1] The meaning of this expression is: the target object under a specific game character.

(2) It can be a combination of the game product ID and its item attribute information.

(3) Label: Each pair and Corresponding to a binary label, 0 means that the target object has not purchased the item under the character attribute information, and 1 means that the target object has purchased the item under the character attribute information. The specific label value can be set independently.

In step S550, the neural network model is trained using the training data set.

In the embodiment of the present invention, during the model training process, the training data set may be and The input is sent to the neural network model, which outputs the predicted value. The loss function is calculated based on the predicted value and the true label. The parameter gradient is calculated with the goal of minimizing the loss function, and the gradient is back-propagated to update the model parameters.

The above process is iterated to continuously update the model parameters and verify the model through the validation set data. The iteration can be stopped when the loss function on the validation set basically no longer decreases. At this time, a model with good fitting and generalization capabilities is obtained.

In the embodiment of the present invention, the loss function may adopt a binary cross entropy loss function loss, which is calculated as follows:

In the above formula, k is a positive integer greater than or equal to 1 and less than or equal to m, and m is a positive integer greater than or equal to 1, which means that there are a total of m pairs in the training data set. and y _k is when the model inputs the kth pair and The corresponding true label is 0 or 1; When the model inputs the kth pair and The predicted value output by the model at this time is a value between [0,1]. The goal of optimization is to minimize the loss. When the predicted value is close to the true value, the loss is small, and when the predicted value deviates from the true value, the loss is large.

In an embodiment of the present invention, a model testing process may also be included. Model testing refers to testing the predictive ability of the model on the trained model using a test data set that does not overlap with the training data set. In order to more intuitively evaluate the predictive ability of the model, the evaluation criteria used on the test data set are the prediction accuracy, recall rate, and AUC (Area Under Curve, the area under the ROC (receiver operating characteristic curve) curve and the coordinate axis). The accuracy rate can represent the precision ability of the model, the recall rate can represent the completeness ability of the model, and the AUC is a combination of the two.

By comparing the precision, recall, and AUC values of different models on the test data set, we can observe the predictive ability of each model.

In the embodiment of the present invention, during the model training process, not only the historical operation records of the target object on the object to be recommended are considered, but also the historical role attribute information of the target object and the item attribute information of the object to be recommended are comprehensively considered, thereby improving the prediction accuracy of the model.

Fig. 6 is a schematic diagram showing the processing process of step S320 in one embodiment shown in Fig. 3. As shown in Fig. 6, in the embodiment of the present invention, the above step S320 may further include the following steps.

In step S321, object attribute information of the target object is obtained.

For example, if the target object is a game player, the object attribute information may be the real name, age, gender, geographical location, historical online time, and other personal portrait information of the game player.

In step S322, the first object data is obtained according to the first role attribute information and the object attribute information.

Specifically, the first character attribute information and the object attribute information can be vectorized to form the first object data. That is, when recommending props to the target object, not only the virtual attribute information of the currently selected game character is included, but also the real attribute information of the target object can be comprehensively considered, thereby further improving the recommendation accuracy.

Fig. 7 is a schematic diagram showing the processing process of step S340 in one embodiment shown in Fig. 3. In the embodiment of the present invention, the neural network model may include a first neural network sub-model, a second neural network sub-model and a third neural network sub-model.

As shown in FIG. 7 , in the embodiment of the present invention, the above step S340 may further include the following steps.

In step S341, an object vector is generated according to the first object data.

In step S342, an item vector of the item to be recommended is generated according to the item attribute information.

In step S343, the object vector and the item vector are processed by the first neural network sub-model to obtain the object preference vector of the first character of the target object and the item characteristic vector of the object to be recommended.

In step S344, the object vector and the item vector are processed by the second neural network sub-model to obtain an interaction relationship vector between the first character of the target object and the item to be recommended.

In step S345, the object preference vector, the item characteristic vector and the interaction relationship vector are processed by the third neural network sub-model to obtain a predicted probability value of the first character of the target object operating the item to be recommended.

In step S346, a first target item recommended to the first character of the target object is determined according to the predicted probability value.

In an embodiment of the present invention, the props in the prop pool can be sorted from large to small according to the predicted probability value of each prop, and the props with a predetermined number (e.g., 10, the specific value is not limited) or a predetermined proportion (e.g., 10%, the specific value is not limited) are selected as the first target items and returned to the game client for display, and the props with the largest predicted probability value are displayed in the first place, and the props with the smallest predicted probability value in the first target item are displayed in the last place, but the present invention is not limited to this. In other embodiments, if the number of all props in the prop pool is small, all props can also be displayed on the game client, but the props are sorted according to the size of the predicted probability value.

In an embodiment of the present invention, the data input to the model can be preprocessed first. The features of each dimension of the data include discrete features (categorical features) and continuous features (dense features). For discrete features, embedding can be used to convert sparse vectors into dense representations. For continuous features, since the data distribution of some continuous features is often a long-tail distribution, the values with large differences under this distribution will interfere with the model. Therefore, in an embodiment of the present invention, the continuous features are first binned and then the data is normalized.

Fig. 8 is a schematic diagram showing the processing process of step S341 in one embodiment shown in Fig. 7. In the embodiment of the present invention, the neural network model may further include a first embedding sub-model, and the first object data may include object discrete features and object continuous features.

As shown in FIG. 8 , in the embodiment of the present invention, the above step S341 may further include the following steps.

In step S3411, the discrete features of the object are processed by the first embedding sub-model to obtain an object embedding vector.

In step S3412, the continuous features of the object are binned to obtain a discrete representation of the object.

In the embodiment of the present invention, any binning method such as equal frequency binning, equal distance binning, chi-square binning, etc. can be adopted, and there is no limitation on this.

In step S3413, the discrete representation of the object is normalized.

In step S3414, the object embedding vector and the normalized object discrete representation are concatenated to generate the object vector.

FIG9 schematically shows a schematic diagram of a first embedding sub-model according to an embodiment of the present invention.

As shown in FIG9 , the first object data Assume that they are 4, 3, …, 0.2, 0.5, …. 4 and 3 are discrete features of the object, and their unique hot codes are 00001 and 0001 respectively. They are input into the embedding layer of the first embedding sub-model, and the dense object embedding vector is output. 0.2 and 0.5 are the object discrete representations of the normalized continuous features of the object, which remain unchanged and are concatenated with the object embedding vector to output the object vector.

Fig. 10 is a schematic diagram showing the processing process of step S342 in one embodiment shown in Fig. 7. In the embodiment of the present invention, the neural network model may further include a second embedding sub-model, and the item attribute information of the item to be recommended may include item discrete features and item continuous features.

As shown in FIG. 10 , in an embodiment of the present invention, the above step S342 may further include the following steps.

In step S3421, the discrete features of the object are processed by the second embedding sub-model to obtain an object embedding vector.

In step S3422, the continuous features of the object are binned to obtain discrete representations of the object.

In step S3423, the discrete representation of the object is normalized.

In step S3424, the item embedding vector and the normalized item discrete representation are concatenated to generate the item vector.

FIG. 11 schematically shows a schematic diagram of a second embedding sub-model according to an embodiment of the present invention.

As shown in Figure 11, item attribute information Assume that 2, 1, ..., 0.4, 0.1, .... 2 and 1 are discrete features of the item, and their unique hot encodings are 00100 and 0100 respectively. They are input into the embedding layer of the second embedding sub-model, and the dense item embedding vector is output. 0.4 and 0.1 are the discrete representations of the normalized continuous features of the item, which remain unchanged and are concatenated with the item embedding vector to output the item vector.

FIG12 schematically shows a structural diagram of a neural network model according to an embodiment of the present invention.

As shown in Figure 12, the first object data is input into the first embedding sub-model to output the object vector; the item attribute information is input into the second embedding sub-model to output the item vector. The object vector is then input into the first neural network sub-model to output the object preference vector and the item characteristic vector; the item vector is input into the second neural network sub-model to output the interaction relationship vector. The object preference vector, item characteristic vector and interaction relationship vector are input into the third neural network sub-model to output the predicted probability value.

FIG13 is a schematic diagram showing the processing process of step S343 shown in FIG7 in one embodiment. In an embodiment of the present invention, the first neural network sub-model may include a first neural network unit and a second neural network unit. The first neural network sub-model is a representation learning module, which can be further divided into two parts: one part is used to learn the vector representation of the target object, and the other part is used to learn the vector representation of the object to be recommended.

As shown in FIG. 13 , in an embodiment of the present invention, the above step S343 may further include the following steps.

In step S3431, the object vector is processed by the first neural network unit to obtain the object preference vector.

In step S3432, the object vector is processed by the second neural network unit to obtain the object feature vector.

FIG14 schematically shows a schematic structural diagram of a first neural network sub-model according to an embodiment of the present invention.

As shown in Figure 14, the object vector Input to the first neural network unit MLP _user , MLP _user includes layers 1, 2 to n1 connected in sequence, n1 is a positive integer greater than or equal to 1, where n1 can be 3 to 5, but the present invention is not limited to this. The first neural network unit MLP _user outputs the object preference vector Vectorize the item Input to the second neural network unit MLP _item , MLP _item includes layers 1, 2 to n2 connected in sequence, n2 is a positive integer greater than or equal to 1, where n2 can be 3 to 5, n1 can be the same as n2 or different, and the present invention is not limited to this. The second neural network unit MLP _item outputs the item feature vector

In the embodiment of FIG14 , the first neural network unit and the second neural network unit are both illustrated by using a multilayer perceptron (MLP). However, in other embodiments, the first neural network unit and/or the second neural network unit may also use other deep learning networks, such as LSTM (Long Short-Term Memory), GRU (Gated Recurrent Unit), etc., and the first neural network unit may use the same deep learning network as the second neural network unit, or a different deep learning network, which is not limited in the present invention. Among them, the multilayer perceptron is a forward-structured neural network that maps a set of input vectors to a set of output vectors. In the embodiment of FIG14 , an MLP learning vector representation is used, that is, the inputs are respectively and Output the corresponding and

Fig. 15 is a schematic diagram showing the processing process of step S344 in one embodiment shown in Fig. 7. As shown in Fig. 15, in the embodiment of the present invention, the above step S344 may further include the following steps.

In step S3441, the object vector and the item vector are concatenated to obtain a concatenated vector.

In step S3442, the concatenated vector is processed by the second neural network sub-model to obtain the interaction relationship vector.

FIG16 schematically shows a schematic diagram of the structure of a second neural network sub-model according to an embodiment of the present invention.

As shown in Figure 16, the object vector and the item vector Cascade (concatenate) to obtain a concatenated vector; input the concatenated vector to the second neural network sub-model MLP _{user_item} , MLP _{user_item} includes layers 1, 2 to n3 connected in sequence, n3 is a positive integer greater than or equal to 1, where n3 can be 3 to 5, n3 can be equal to n1 or n2, or it can be different, and the present invention is not limited to this. The second neural network sub-model MLP _{user_item} outputs the interaction relationship vector

In the embodiment of the present invention, the second neural network sub-model MLP _{user_item} can also be called a relation learning module. In the representation learning module, the representations of the target object and the item to be recommended are learned separately, while the relation learning module is designed to learn the complex interactive relationship between the target object and the item to be recommended. and Concatenate them, and then send the concatenated vector to MLP _{user_item} . Compared with the use of linear functions such as vector inner product in related technologies to characterize the relationship between the target object and the item to be recommended, the nonlinear structure of MLP is more conducive to describing the complex process from representation to result.

In other embodiments, in addition to MLP, the second neural network sub-model may also use other neural networks or deep learning networks, such as LSTM, GRU, etc. The first neural network sub-model and the second neural network sub-model may both use MLP, or may both use other identical neural networks, or may respectively use different neural networks.

Deep learning, as a technology suitable for processing large-scale complex data and with strong expressiveness, is applied here in the recommendation system. The embodiment of the present invention designs a collaborative filtering network structure based on deep learning in the game scenario, and uses a neural network instead of the traditional vector inner product to calculate the interactive relationship between users and products.

Fig. 17 is a schematic diagram showing the processing process of step S345 in one embodiment shown in Fig. 7. As shown in Fig. 17, in the embodiment of the present invention, the above step S345 may further include the following steps.

In step S3451, the object preference vector is multiplied by the item characteristic vector to obtain a dot product vector.

In step S3452, the dot product vector and the interaction relationship vector are concatenated to obtain a concatenated vector.

In step S3453, the concatenated vector is processed by the third neural network sub-model to obtain the predicted probability value.

In the embodiment of the present invention, the third neural network sub-model can also be called a joint prediction layer. After the representation learning module and the relationship learning module, it is necessary to predict the purchase probability of the target object for the game product. The joint prediction layer combines the outputs of the representation learning module and the relationship learning module to calculate the final prediction result. For example, the calculation method of the joint prediction layer can be as follows:

In the above formula, ⊙ represents the multiplication of vector elements. represents vector concatenation, the selected function f can be a single fully connected layer plus a softmax function, and the output is a two-dimensional vector of the form The first digit represents the probability value that the user does not purchase the product, and the second digit represents the probability value that the user purchases the product, which is the predicted probability value mentioned above. The sum of the two probabilities is 1.

The following is an example of applying the solution provided in the above embodiment to a prop recommendation scenario of an RPG game. In this game, users can choose game characters, which are associated with a variety of attributes, such as character cultivation, skill, etc. (the value will increase with the length of online time); users can enter the game mall to buy props needed by the character, and props are also associated with multiple types of attributes. The purpose of recommendation in this scenario is to recommend one or more props that the user is most likely to be interested in based on the attribute information of the user (the user's real gender and other attribute information can be comprehensively considered, not limited to the example user ID), character, and prop.

This RPG game is a martial arts theme MMORPG (Massive Multiplayer Online Role-Playing Game), including eight sects, including Taibai, Shenwei, Tangmen, Beggars, Zhenwu, Tianxiang, Wudu, Shaolin and other sects; it also includes hundreds of professions in the rivers and lakes, and the scale of the guilds such as constables, escorts, hunters, rangers, killers, musicians, and literati is the largest, followed by hanging eyes, merchants, and market professionals. Users can download and install the game client. After the game client is installed, click the game icon on the desktop to run the game launcher. Click "Enter the Game" on the right side of the launcher to enter the game account login interface. Enter the game account and password on this interface and click "Enter the Game". View "More Servers" to open the server list, select the game area and server to log in (you can also directly select the server you recently logged in on the right), and click "Start Game" to start your journey in the rivers and lakes.

FIG18 schematically shows an interface diagram of a martial arts introduction according to an embodiment of the present invention.

As shown in Figure 18, the user first creates a character. The first step is to select a school. If it is the first time for the user to log in to the game, the game account does not have any character on the server, and the game school selection interface will be displayed. You can select any open school in this interface to create the user's character. Click any open school to enter the school introduction interface. The currently selected school and character appearance can be seen in the middle of the interface. The right side of the interface lists the combat characteristics of the school. The gender of the character can be switched below. Click the right mouse button to rotate the character to view the image. The left side can quickly switch the character school.

19 and 20 schematically illustrate a schematic diagram of an interface for customizing character details according to an embodiment of the present invention.

As shown in Figure 19, click "Customize Details" to further customize the character details. In this interface, you can see more details of the character and fine-tune each item. According to the natural distribution of human facial bones and muscles, there are more than 200 adjustable parameters on 48 bones. At the same time, as shown in Figure 20, it also provides previews of multiple (4 as an example here) facial expressions and multiple (5 as an example here) outfits to try on. After entering the character name at the bottom of the interface, click "Create Complete" to create the user's character. After creating the character, select the created character and click "Start Game" below the character as shown in Figure 21 to enter the game scene as shown in Figure 22.

As shown in Figure 22, there are the side shortcut bar, chat interface, character status, bottom shortcut bar, function buttons (character and mall), mini map, and task gameplay guide.

The function buttons on the game interface are the entrances to the most important functional modules of the game. Among them, the roles include: attributes, meridians, mental skills, equipment, etc. The identities include: escort, assassin, ranger, hunter, musician, scholar, and constable. The three roles of hanging eyes, merchants, and market people are not open yet. The mall includes: common, appearance, mount pendants, treasures, member exclusive, Tianshang points, bound coupons, purchase membership, coupon consignment, and coupon recharge.

The character status is used to view the current character's Qi, blood, internal energy, concentration, killing intent, and the unique moves of each sect.

The shortcut bar at the bottom corresponds to QERTG and 0-9, a total of 15 keys. The up and down buttons on the right can be used to switch. The shortcut bar can be used to place props and skills, and they can be used quickly by pressing the corresponding buttons. The keys on the shortcut bar can be changed in the system settings.

The side shortcut bar corresponds to F1-F10, Ctrl0-Ctrl9 and HVCXZ, a total of 25 keys, where you can place items and skills, and use them quickly by pressing the corresponding keys. The keys on the shortcut bar can be changed in the system settings.

The chat interface can be used to display the current chat information with other players in the game and system messages. In the upper part of the chat window, you can select the channel to speak and enter the content of the speech.

In the player information, you can view the current character's avatar, sect, name, and level information.

The mini-map shows the surrounding map information of the player's current environment. The button on the upper left corner is the calendar, which can display the current area, date, time and weather; the button on the upper right corner is the mailbox; the buttons on the lower right corner are: zoom in, zoom out, view world map, view large map.

The mission gameplay guide is used to track the current story mission/game activity information. Click the link above to be directed to the target point, and click the story/game tab above to switch to view the story mission and gameplay activity information.

In this game, experience is closely related to the improvement of character level, that is, character experience. The experience value is displayed at the top of the game client, with a yellow progress bar showing the current experience and the experience required for the next level. In addition to completing the main quest, experience can also be obtained through sect meditation, guild daily, bandit fighting and other supplementary gameplay. You can open the "Daily Must-Do" on the left side of the small map in the upper right corner of the game interface to view the daily completion status.

FIG. 23 schematically shows an interface diagram of role attributes according to an embodiment of the present invention.

As shown in Figure 23, the character attributes may include: sect (e.g., Tang Sect), title, merit value, killing, cultivation, vitality, etc., among which cultivation can be used to cultivate the meridians and mental methods of the character. You can click to select to view all attributes, meridian attributes, mental method attributes, and equipment attributes. All attributes may include basic attributes and combat attributes. Basic attributes may include Qi and blood, internal breath, concentration, skill, strength, root bones, Qi, insight, and body skills. Combat attributes may include external attack, internal attack, concentration attack, hit rate, critical rate, critical damage, external defense, internal defense, concentration defense, blocking, and tenacity.

FIG. 24 schematically shows an interface diagram of a game mall according to an embodiment of the present invention.

As shown in Figure 24, you can choose to view all props, equipment props, Bian stone props, mind props, and identity props in the game mall. All props can include purple mind annotations, Xuantian stone mother, Tibetan antelope leather, phoenix dance picture, pearls, cultivation pills, casting god orders, jade glass, gold wire spikes, and dragon cinnabar. Here, the display cycle for product attributes is set to 7 days, and the order of props is updated every 7 days according to the changes in the user's role attributes.

FIG. 25 schematically shows a block diagram of an information processing device according to an embodiment of the present invention.

As shown in FIG. 25 , the information processing device 2500 provided by the embodiment of the present invention may include: a first object role determination module 2510 , a first object data acquisition module 2520 , an item attribute information acquisition module 2530 and a first target item determination module 2540 .

Among them, the first object role determination module 2510 can be configured to determine the first role of the target object. The first object data acquisition module 2520 can be configured to obtain the first object data of the target object according to the first role attribute information of the first role. The item attribute information acquisition module 2530 can be configured to obtain the item attribute information of the item to be recommended. The first target item determination module 2540 can be configured to process the first object data and the item attribute information through a neural network model, and determine the first target item recommended to the first role of the target object from the items to be recommended.

In an exemplary embodiment, the information processing device 2500 may also include: a second object role determination module, which can be configured to determine the second role of the target object; a second object data acquisition module, which can be configured to obtain the second object data of the target object based on the second role attribute information of the second role; a second target item determination module, which can be configured to process the second object data and the item attribute information through the neural network model, and determine the second target item recommended to the second role of the target object from the items to be recommended.

In an exemplary embodiment, the information processing device 2500 may also include: a historical role acquisition module, which can be configured to obtain the historical role of the target object; a historical object data acquisition module, which can be configured to obtain the historical object data of the target object based on the historical role attribute information of the historical role; a historical operation record acquisition module, which can be configured to obtain the historical operation record of the target object on the object to be recommended; a training set generation module, which can be configured to determine the labels of the historical object data and the item attribute information based on the historical operation record to generate a training data set; and a model training module, which can be configured to train the neural network model using the training data set.

In an exemplary embodiment, the neural network model may include a first neural network sub-model, a second neural network sub-model, and a third neural network sub-model. The first target item determination module 2540 may include: an object vector generation unit, which may be configured to generate an object vector according to the first object data; an item vector generation unit, which may be configured to generate an item vector of the item to be recommended according to the item attribute information; an object item characteristic vector acquisition unit, which may be configured to process the object vector and the item vector through the first neural network sub-model to obtain an object preference vector of the first role of the target object and an item characteristic vector of the item to be recommended; an interaction relationship vector acquisition unit, which may be configured to process the object vector and the item vector through the second neural network sub-model to obtain an interaction relationship vector between the first role of the target object and the item to be recommended; a prediction probability acquisition unit, which may be configured to process the object preference vector, the item characteristic vector, and the interaction relationship vector through the third neural network sub-model to obtain a prediction probability value of the first role of the target object operating the item to be recommended; and a first target item determination unit, which may be configured to determine the first target item recommended to the first role of the target object according to the prediction probability value.

In an exemplary embodiment, the first neural network sub-model may include a first neural network unit and a second neural network unit. The object item characteristic vector obtaining unit may include: an object preference vector obtaining sub-unit, which may be configured to process the object vector through the first neural network unit to obtain the object preference vector; and an item characteristic vector obtaining sub-unit, which may be configured to process the item vector through the second neural network unit to obtain the item characteristic vector.

In an exemplary embodiment, the interaction relationship vector obtaining unit may include: a splicing vector obtaining subunit, which can be configured to splice the object vector with the item vector to obtain a splicing vector; and an interaction relationship vector obtaining subunit, which can be configured to process the splicing vector through the second neural network sub-model to obtain the interaction relationship vector.

In an exemplary embodiment, the predicted probability obtaining unit may include: a dot product vector obtaining subunit, which can be configured to multiply the object preference vector with the item characteristic vector to obtain a dot product vector; a concatenation vector obtaining subunit, which can be configured to concatenate the dot product vector with the interaction relationship vector to obtain a concatenation vector; and a predicted probability obtaining subunit, which can be configured to process the concatenation vector through the third neural network submodel to obtain the predicted probability value.

In an exemplary embodiment, the neural network model may further include a first embedding sub-model, and the first object data may include object discrete features and object continuous features. The object vector generation unit may include: an object embedding vector acquisition sub-unit, which may be configured to process the object discrete features through the first embedding sub-model to obtain an object embedding vector; an object discrete representation acquisition sub-unit, which may be configured to bin the object continuous features to obtain an object discrete representation; an object normalization processing sub-unit, which may be configured to normalize the object discrete representation; and an object vector generation sub-unit, which may be configured to concatenate the object embedding vector and the normalized object discrete representation to generate the object vector.

In an exemplary embodiment, the neural network model may further include a second embedding sub-model, and the item attribute information of the item to be recommended may include item discrete features and item continuous features. The item vector generation unit may include: an item embedding vector acquisition sub-unit, which may be configured to process the item discrete features through the second embedding sub-model to obtain an item embedding vector; an item discrete representation acquisition sub-unit, which may be configured to bin the item continuous features to obtain an item discrete representation; an item normalization processing sub-unit, which may be configured to normalize the item discrete representation; and an item vector generation sub-unit, which may be configured to concatenate the item embedding vector and the normalized item discrete representation to generate the item vector.

In an exemplary embodiment, the first object data obtaining module 2520 may include: an object attribute information obtaining unit, which may be configured to obtain object attribute information of the target object; and a first object data obtaining unit, which may be configured to obtain the first object data according to the first role attribute information and the object attribute information.

In an exemplary embodiment, the first role attribute information of the first role may include identity information, current level information and/or current operation information of the first role.

The specific implementation of each module, unit and sub-unit in the information processing device provided by the embodiment of the present invention can refer to the content of the above-mentioned information processing method, which will not be repeated here.

It should be noted that, although several modules, units and subunits of the equipment for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to an embodiment of the present invention, the features and functions of two or more modules, units and subunits described above can be concretized in one module, unit and subunit. On the contrary, the features and functions of one module, unit and subunit described above can be further divided into being concretized by multiple modules, units and subunits.

Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present invention can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a touch terminal, or a network device, etc.) to execute the method according to the implementation of the present invention.

Those skilled in the art will readily appreciate other embodiments of the present invention after considering the specification and practicing the invention disclosed herein. This application is intended to cover any variations, uses or adaptations of the present invention that follow the general principles of the present invention and include common knowledge or customary techniques in the art that are not disclosed by the present invention. The specification and examples are to be considered exemplary only, and the true scope and spirit of the present invention are indicated by the following claims.

It should be understood that the present invention is not limited to the exact construction that has been described above and shown in the drawings and that various modifications and changes may be made without departing from the scope thereof. The scope of the present invention is limited only by the appended claims.

Claims (14)

1. An information processing method, comprising: Determine the first role of the target object; Acquire object attribute information of the target object, wherein the object attribute information includes any one or more of the real name, age, gender, geographical location, and historical online time of the target object; Obtaining first object data of the target object according to the first role attribute information of the first role and the object attribute information; Obtaining item attribute information of the item to be recommended, the item to be recommended including props, the item attribute information including any one or more of the category to which each prop belongs, the function that can be completed, the identity of the corresponding character, the level of the corresponding character, and the operation of the corresponding character; Generate an object vector according to the first object data, process the object vector through a first neural network unit to obtain an object preference vector of a first role of the target object; generate an item vector of the object to be recommended according to the item attribute information, process the item vector through a second neural network unit to obtain an item characteristic vector of the object to be recommended; the first neural network sub-model includes the first neural network unit and the second neural network unit; The object vector and the item vector are concatenated to obtain a concatenated vector, and the concatenated vector is processed by a second neural network sub-model to obtain an interaction relationship vector between the first role of the target object and the item to be recommended, wherein the second neural network sub-model adopts MLP, LSTM or GRU; Multiplying the object preference vector and the item characteristic vector to obtain a dot product vector; Concatenate the dot product vector and the interaction relationship vector to obtain a concatenated vector; The concatenated vector is processed by a third neural network sub-model to obtain a predicted probability value of the first character of the target object operating the object to be recommended; the neural network model includes the first neural network sub-model, the second neural network sub-model and the third neural network sub-model; A first target item recommended to a first character of the target object is determined according to the predicted probability value. 2. The method according to claim 1, further comprising: determining a second role of the target object; Obtaining second object data of the target object according to the second role attribute information of the second role; The second object data and the item attribute information are processed by the neural network model to determine a second target item recommended to the second character of the target object from the items to be recommended. 3. The method according to claim 1, further comprising: Obtaining the historical role of the target object; Obtaining historical object data of the target object according to the historical role attribute information of the historical role; Obtaining historical operation records of the target object on the item to be recommended; Determining labels of the historical object data and the item attribute information according to the historical operation records to generate a training data set; The neural network model is trained using the training data set. 4. The method according to claim 1, wherein the first object data comprises object discrete features and object continuous features; wherein generating an object vector according to the first object data comprises: Processing the discrete features of the object through a first embedding sub-model to obtain an object embedding vector; Binning the continuous features of the object to obtain a discrete representation of the object; Normalizing the discrete representation of the object; The object embedding vector and the normalized object discrete representation are concatenated to generate the object vector. 5. The method according to claim 1, characterized in that the item attribute information of the item to be recommended includes item discrete features and item continuous features; wherein generating the item vector of the item to be recommended according to the item attribute information comprises: Processing the discrete features of the item through a second embedding sub-model to obtain an item embedding vector; Binning the continuous features of the object to obtain a discrete representation of the object; Normalizing the discrete representation of the object; The item embedding vector and the normalized item discrete representation are concatenated to generate the item vector. 6. The method according to claim 1 is characterized in that the first role attribute information of the first role includes identity information, current level information and/or current operation information of the first role. 7. An information processing device, comprising: A first object role determination module, configured to determine a first role of a target object; A first object data acquisition module is configured to acquire object attribute information of the target object, and acquire first object data of the target object according to the first role attribute information of the first role and the object attribute information, wherein the object attribute information includes any one or more of the real name, age, gender, geographical location, and historical online time of the target object; An item attribute information obtaining module, configured to obtain item attribute information of an item to be recommended, wherein the item to be recommended includes a prop, and the item attribute information includes any one or more of the category to which each prop belongs, the function that can be completed, the identity of the corresponding character, the level of the corresponding character, and the operation of the corresponding character; A first target item determination module is configured to process the first object data and the item attribute information through a neural network model, and determine a first target item recommended to the first character of the target object from the items to be recommended; The neural network model includes a first neural network sub-model, a second neural network sub-model and a third neural network sub-model; wherein the first target item determination module includes: an object vector generating unit, configured to generate an object vector according to the first object data; An item vector generating unit, configured to generate an item vector of the item to be recommended according to the item attribute information; an object item characteristic vector obtaining unit, configured to process the object vector and the item vector through the first neural network sub-model to obtain an object preference vector of the first role of the target object and an item characteristic vector of the to-be-recommended item; an interactive relationship vector obtaining unit, configured to process the object vector and the item vector through the second neural network sub-model to obtain an interactive relationship vector between the first role of the target object and the item to be recommended, wherein the second neural network sub-model adopts MLP, LSTM or GRU; a prediction probability obtaining unit, configured to process the object preference vector, the item characteristic vector, and the interaction relationship vector through the third neural network sub-model to obtain a prediction probability value of the first character of the target object operating the to-be-recommended item; A first target item determining unit, configured to determine a first target item recommended to a first character of the target object according to the predicted probability value; The first neural network sub-model includes a first neural network unit and a second neural network unit; wherein the object item characteristic vector obtaining unit includes: an object preference vector obtaining subunit, configured to process the object vector through the first neural network unit to obtain the object preference vector; an item characteristic vector obtaining subunit, configured to process the item vector through the second neural network unit to obtain the item characteristic vector; The interactive relationship vector obtaining unit comprises: a splicing vector obtaining subunit, configured to splice the object vector with the item vector to obtain a splicing vector; an interaction relationship vector obtaining subunit, configured to process the splicing vector through the second neural network submodel to obtain the interaction relationship vector; The prediction probability obtaining unit comprises: a dot product vector obtaining subunit, configured to multiply the object preference vector and the item characteristic vector to obtain a dot product vector; A concatenated vector obtaining subunit, configured to concatenate the dot product vector and the interaction relationship vector to obtain a concatenated vector; The prediction probability obtaining subunit is configured to process the concatenated vector through the third neural network sub-model to obtain the prediction probability value. 8. The device according to claim 7, further comprising: A second object role determination module, configured to determine a second role of the target object; A second object data obtaining module, configured to obtain second object data of the target object according to the second role attribute information of the second role; The second target item determination module is configured to process the second object data and the item attribute information through the neural network model, and determine the second target item recommended to the second character of the target object from the items to be recommended. 9. The device according to claim 7, further comprising: A historical role obtaining module, configured to obtain the historical role of the target object; A historical object data acquisition module, configured to acquire the historical object data of the target object according to the historical role attribute information of the historical role; A historical operation record obtaining module, configured to obtain the historical operation record of the target object on the object to be recommended; A training set generation module, configured to determine labels of the historical object data and the item attribute information according to the historical operation records to generate a training data set; The model training module is configured to train the neural network model using the training data set. 10. The device according to claim 7, characterized in that the neural network model further comprises a first embedding sub-model, the first object data comprises object discrete features and object continuous features; wherein the object vector generation unit comprises: an object embedding vector obtaining subunit, configured to process the discrete features of the object through the first embedding submodel to obtain an object embedding vector; An object discrete representation obtaining subunit is configured to perform binning processing on the continuous features of the object to obtain a discrete representation of the object; An object normalization processing subunit, configured to perform normalization processing on the discrete representation of the object; The object vector generating subunit is configured to concatenate the object embedding vector and the normalized object discrete representation to generate the object vector. 11. The device according to claim 7, characterized in that the neural network model further comprises a second embedding sub-model, the item attribute information of the item to be recommended comprises item discrete features and item continuous features; wherein the item vector generation unit comprises: an item embedding vector obtaining subunit, configured to process the item discrete features through the second embedding submodel to obtain an item embedding vector; An item discrete representation obtaining subunit is configured to perform binning processing on the continuous features of the item to obtain the item discrete representation; An item normalization processing subunit, configured to perform normalization processing on the discrete representation of the item; The item vector generation subunit is configured to concatenate the item embedding vector and the normalized item discrete representation to generate the item vector. 12. The device according to claim 7 is characterized in that the first role attribute information of the first role includes identity information, current level information and/or current operation information of the first role. 13. An electronic device, comprising: one or more processors; A storage device configured to store one or more programs, when the one or more programs are executed by the one or more processors, enables the one or more processors to implement the information processing method according to any one of claims 1 to 6. 14. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the information processing method according to any one of claims 1 to 6.