Network request processing method and apparatus, device, storage medium, and program product

Abstract

The application provides an artificial intelligence-based network request processing method and device, electronic equipment, computer readable storage medium and computer program product, and relates to the technical field of artificial intelligence. The method comprises the following steps: obtaining relevant information of a target object; performing feature extraction processing on the relevant information of the target object to obtain object features of the target object; performing classification processing on the relevant information of the target object based on the object features of the target object to obtain an information category to which the relevant information belongs; performing meta-learning-based point of interest prediction processing based on the information category to which the relevant information belongs to obtain a target point of interest to be passed through by the target object; and applying a network service mode matched with the target point of interest to the target object.

Description

Technical Field

[0001] This application relates to artificial intelligence technology, and more particularly to a network request processing method, apparatus, electronic device, computer-readable storage medium, and computer program product based on artificial intelligence. Background Technology

[0002] Artificial Intelligence (AI) is a comprehensive technology within computer science that studies the design principles and implementation methods of various intelligent machines, enabling them to possess perception, reasoning, and decision-making capabilities. AI technology is a multidisciplinary field encompassing a wide range of areas, including natural language processing and machine learning / deep learning. With technological advancements, AI will be applied in more fields and play an increasingly important role.

[0003] Currently, there is a lack of effective solutions for network request processing based on points of interest (POIs). These solutions primarily rely on manually defined rules to identify the next POI along the trajectory of a user's movement and apply network service modes upon reaching that POI. However, this approach often results in inaccurate identification of the next POI, wasting significant computational resources. Summary of the Invention

[0004] This application provides a network request processing method, apparatus, electronic device, computer-readable storage medium, and computer program product based on artificial intelligence, which can automatically and accurately predict the points of interest that a movement trajectory will pass through.

[0005] The technical solution of this application embodiment is implemented as follows:

[0006] This application provides an artificial intelligence-based network request processing method, including:

[0007] Acquire a set of feature data and movement trajectory of the target object, wherein the movement trajectory includes at least one historical point of interest along a path;

[0008] The feature data set and movement trajectory of the target object are subjected to feature extraction processing to obtain the object features of the target object in relation to the movement trajectory;

[0009] Based on the object characteristics of the target object in relation to the action trajectory, the action trajectory of the target object is classified to obtain the trajectory category to which the action trajectory belongs;

[0010] Based on the trajectory category to which the action trajectory belongs, meta-learning-based interest point prediction processing is performed to obtain the target interest points that the action trajectory will pass through.

[0011] apply a network service mode matched with the target point of interest to the target object.

[0012] The embodiment of the application provides a network request processing device based on artificial intelligence, comprising:

[0013] An acquisition module is configured to acquire a feature data set of a target object and a movement trajectory, wherein the movement trajectory comprises at least one historical point of interest of a route;

[0014] A feature extraction module is configured to perform feature extraction processing on the feature data set of the target object and the movement trajectory to obtain object features of the target object for the movement trajectory.

[0015] A classification module is configured to perform trajectory classification processing on the movement trajectory of the target object based on the object features of the target object for the movement trajectory to obtain a trajectory category to which the movement trajectory belongs.

[0016] A prediction module is configured to perform meta-learning-based point of interest prediction processing based on the trajectory category to which the movement trajectory belongs to obtain a target point of interest to be passed through by the movement trajectory.

[0017] An application module is configured to apply a network service mode matched with the target point of interest to the target object.

[0018] In the above technical solution, the feature extraction module is further configured to perform fusion processing on the feature data set of the target object and the movement trajectory to obtain fusion information of the target object.

[0019] The feature extraction processing is performed on the fusion information of the target object to obtain the object features of the target object for the movement trajectory.

[0020] In the above technical solution, the feature extraction processing is implemented through a feature extraction model, and the feature extraction model comprises a plurality of cascaded feature extraction layers.

[0021] The feature extraction result of the first feature extraction layer is output to a subsequent cascaded feature extraction layer, and the feature extraction processing and the feature extraction result output are continued through the subsequent cascaded feature extraction layer until the last feature extraction layer.

[0022] The feature extraction result output by the last feature extraction layer is taken as the object features of the target object for the movement trajectory.

[0023] In the technical solution, the feature extraction module is further configured to perform the following processing through the jth feature extraction layer of the plurality of cascaded feature extraction layers:

[0024] performing random convolution processing on the feature extraction result of the (j-1)th feature extraction layer to obtain the jth deep layer feature;

[0025] performing pooling processing on the jth deep layer feature to obtain the feature extraction result of the jth feature extraction layer;

[0026] outputting the feature extraction result of the jth feature extraction layer through the jth feature extraction layer;

[0027] wherein j is an increasing natural number and the value range is 1 < j ≤ N, N is the number of the feature extraction layers, and N is a positive integer greater than 1.

[0028] In the technical solution, the classification module is further configured to perform time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a time sequence feature of the action trajectory;

[0029] performing mapping processing on the time sequence feature of the action trajectory to obtain a trajectory category to which the action trajectory belongs.

[0030] In the technical solution, the classification module is further configured to perform forward time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a forward time sequence feature of the action trajectory;

[0031] performing reverse time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a reverse time sequence feature of the action trajectory;

[0032] performing splicing processing on the forward time sequence feature and the reverse time sequence feature to obtain the time sequence feature of the action trajectory.

[0033] In the technical solution, the classification module is further configured to perform forward screening processing on the object feature based on the (i-1)th forward time sequence feature based on a forgetting gate to obtain a forward output feature of the forgetting gate;

[0034] performing forward updating processing on the object feature based on the (i-1)th forward time sequence feature based on an input gate to obtain a forward output feature of the input gate;

[0035] performing forward mapping processing on the object feature based on the (i-1)th forward time sequence feature based on an output gate to obtain a forward output feature of the output gate;

[0036] performing nonlinear mapping processing on the forward output feature of the forgetting gate, the forward output feature of the input gate, and the forward output feature of the output gate to obtain an i th forward time sequence feature;

[0037] taking a set of M forward time sequence features as the forward time sequence features of the action trajectory;

[0038] wherein i is an increasing natural number and takes a value range of 1 < i ≤ M, M is a feature quantity of the object feature, and M is a positive integer greater than 1.

[0039] In the technical solution, the classification module is further configured to perform forgetting gate-based reverse screening processing on the object feature based on the e+1 th reverse time sequence feature to obtain a reverse output feature of the forgetting gate.

[0040] performing input gate-based reverse update processing on the object feature based on the e+1 th reverse time sequence feature to obtain a reverse output feature of the input gate;

[0041] performing output gate-based reverse mapping processing on the object feature based on the e+1 th reverse time sequence feature to obtain a reverse output feature of the output gate;

[0042] performing nonlinear mapping processing on the reverse output feature of the forgetting gate, the reverse output feature of the input gate, and the reverse output feature of the output gate to obtain an e th reverse time sequence feature;

[0043] taking a set of M reverse time sequence features as the reverse time sequence features of the action trajectory;

[0044] wherein e is a decreasing natural number and takes a value range of 1 ≤ e ≤ M-1, M is a feature quantity of the object feature, and M is a positive integer greater than 1.

[0045] In the technical solution, the interest point prediction processing is implemented through a meta-learning model, and a training process of the meta-learning model includes: constructing a query set and a support set based on a plurality of action trajectory samples, the support set includes at least one support subset, each support subset includes at least one first action trajectory sample in the plurality of action trajectory samples, different support subsets include action trajectory samples of different trajectory categories, and the query set includes at least one second action trajectory sample in the plurality of action trajectory samples.

[0046] performing the following processing through the initialized meta-learning model:

[0047] performing interest point prediction processing based on meta-learning on the first action trajectory sample included in the support subset to obtain a first predicted interest point to be passed through by the first action trajectory sample;

[0048] perform meta-learning-based point of interest prediction processing on a second action trajectory sample included in the query set to obtain a second predicted point of interest to be passed by the second action trajectory sample;

[0049] construct a loss function of the meta-learning model based on the first predicted point of interest, a point of interest label corresponding to the first action trajectory sample, the second predicted point of interest, and a point of interest label corresponding to the second action trajectory sample;

[0050] update parameters of the meta-learning model based on the loss function, and use the updated parameters of the meta-learning model as parameters of the trained meta-learning model.

[0051] In the technical solution, the application module is further configured to query a correspondence between different points of interest and candidate network service modes based on the target point of interest, and use a queried candidate network service mode as a network service mode matched with the target point of interest.

[0052] display recommendation information of the network service mode;

[0053] In response to an application operation for the network service mode based on the recommendation information, apply the network service mode to a terminal corresponding to the target object.

[0054] In the technical solution, the application module is further configured to query a correspondence between different points of interest and candidate network service modes based on the target point of interest, and use a queried candidate network service mode as a network service mode matched with the target point of interest.

[0055] apply the network service mode matched with the target point of interest to a terminal corresponding to the target object.

[0056] In the technical solution, the application module is further configured to obtain a distance from a terminal corresponding to the target object to the target point of interest.

[0057] When the distance is less than a distance threshold, apply the network service mode matched with the target point of interest to a terminal corresponding to the target object.

[0058] In the technical solution, when the network service mode matched with the target point of interest is multiple, the application module is further configured to obtain a target time period when the target point of interest is reached.

[0059] determine a network service mode matched with the target time period from the multiple network service modes matched with the target point of interest;

[0060] apply the network service mode matched with the target time period to a terminal corresponding to the target object.

[0061] The embodiment of the present application provides an electronic device for network request processing, and the electronic device comprises:

[0062] a memory for storing executable instructions;

[0063] a processor for executing the executable instructions stored in the memory, so as to realize the network request processing method based on artificial intelligence provided by the embodiment of the present application.

[0064] The embodiment of the present application provides a computer readable storage medium, which stores executable instructions, and is used for causing a processor to execute the network request processing method based on artificial intelligence provided by the embodiment of the present application.

[0065] The embodiment of the present application provides a computer program product, comprising a computer program or instructions, characterized by the computer program or instructions being executed by a processor to realize the network request processing method based on artificial intelligence provided by the embodiment of the present application.

[0066] The embodiment of the present application has the following beneficial effects:

[0067] Based on the object feature of the target object for the action trajectory, the action trajectory of the target object is classified, the trajectory category to which the action trajectory belongs is obtained, and the interest point prediction is performed based on the trajectory category to which the action trajectory belongs, so that the target interest point to be passed through by the action trajectory is automatically and accurately obtained, and the network service mode application is performed based on the accurate target interest point, so that the accuracy of the network request processing is improved, and the related communication resources and computing resources are saved. BRIEF DESCRIPTION OF DRAWINGS

[0068] Figure 1 FIG. 1 is a schematic diagram of an application scenario of a network request processing system provided by the embodiment of the present application;

[0069] Figure 2 FIG. 3 is a structural schematic diagram of an electronic device for network request processing provided by the embodiment of the present application;

[0070] Figures 3A-3C FIG. 5 is a flow schematic diagram of the network request processing method based on artificial intelligence provided by the embodiment of the present application;

[0071] Figure 4 FIG. 7 is a schematic diagram of a WiFi enhanced package provided by the embodiment of the present application;

[0072] Figure 5 FIG. 9 is a schematic diagram of a WiFi protection plan provided by the embodiment of the present application;

[0073] Figure 6A FIG. 11 is a schematic diagram of video acceleration provided by the embodiment of the present application;

[0074] Figure 6B is a schematic diagram of game acceleration provided by an embodiment of the present application;

[0075] Figure 7 is an architecture diagram of the network request processing method based on artificial intelligence provided by an embodiment of the present application;

[0076] Figure 8 is a flowchart of the network request processing method based on artificial intelligence provided by an embodiment of the present application;

[0077] Figure 9 is an aggregation schematic diagram provided by an embodiment of the present application;

[0078] Figure 10 is a feature extractor and a BLSTM algorithm framework provided by an embodiment of the present application;

[0079] Figure 11 is a convolution schematic diagram provided by an embodiment of the present application;

[0080] Figure 12 is a pooling schematic diagram provided by an embodiment of the present application;

[0081] Figure 13 is a bidirectional long short-term memory network structure diagram provided by an embodiment of the present application;

[0082] Figure 14 is a meta-learning model schematic diagram provided by an embodiment of the present application;

[0083] Figure 15 is a trajectory schematic diagram provided by an embodiment of the present application;

[0084] Figure 16 is a marker schematic diagram provided by an embodiment of the present application;

[0085] Figure 17 is an effect comparison diagram provided by an embodiment of the present application. DETAILED DESCRIPTION

[0086] In order to make the purpose, technical scheme and advantages of the present application clearer, the present application will be described in further detail below with reference to the drawings, and the described embodiments should not be regarded as limiting the present application. All other embodiments obtained by those skilled in the art without creative labor fall within the scope of protection of the present application.

[0087] In the following description, the terms "first\second" are merely to distinguish similar objects, and do not represent a specific order of the objects. It can be understood that the "first\second" can be interchanged in a specific order or sequence as allowed, so that the embodiments described herein can be implemented in an order other than that illustrated or described herein.

[0088] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the description herein is for describing the embodiments of the application only and is not intended to be limiting of the application.

[0089] The relevant data collection process in the embodiments of the application should strictly comply with the requirements of relevant laws and regulations, obtain the informed consent or separate consent of the personal information subject, and within the scope of authorization of laws and regulations and the personal information subject, carry out subsequent data use and processing.

[0090] Before further detailing the embodiments of the application, the terms and phrases involved in the embodiments of the application are explained, and the terms and phrases involved in the embodiments of the application are applicable to the following explanations.

[0091] 1) In response to: used to represent the conditions or states on which the operations performed depend, when the dependent conditions or states are met, one or more operations performed can be real-time or have a set delay; in the absence of special instructions, there is no restriction on the execution order of multiple operations performed.

[0092] 2) Feature data set: as an effective tool for sketching the target object, the contact object appeal and the design direction. The feature data set has been widely used in various fields. In the actual operation process, the attributes, behaviors and expectations of the target object are often connected as virtual representatives of the actual object in the most simple and life-like language.

[0093] 3) Network service mode: a mode for network service (software module running on the network, service-oriented, distributed program-based), applying the network service mode to the terminal corresponding to the target object can optimize the network of the terminal to obtain fast and secure network. The network service mode in the embodiments of the application includes network enhancement mode (such as wireless compatibility authentication (WiFi) enhancement application), network protection mode (WiFi protection application) and network application acceleration mode (such as application acceleration application).

[0094] 4) Point of Interest (POI): represents a check-in point of a target object (e.g., a real target object, a virtual machine program). In a geographic information system, a POI can be a house, a shop, a mailbox, a bus stop, etc. Each POI contains four aspects of information, namely, name, category, coordinate, and classification. Comprehensive POI information is essential information for a rich map, and timely POI information points can remind the object of the branch of the road condition and the detailed information of the surrounding buildings, and can also facilitate the query of various places.

[0095] 5) Object movement mode: according to the spatial geographic data of the object obtained by the terminal and other Internet of Things devices, some travel habits of the object are analyzed, so as to provide corresponding services for the object. For example, trajectory and user linking (TUL, Trajectory User Linking), which judges which object mode the trajectory belongs to, the purpose is to correctly classify the trajectory into the correct object mode; next POI prediction, the embodiment of the application regards POI prediction as a multi-classification problem, given an object u with historical trajectory and a recently visited POI sequence Tj=(cn+1,cn+1,…,cn+k-1), train a model M to predict the next POI of the object u, that is, cn+k.

[0096] 6) Deep learning (DL): derived from the research of artificial neural network, multi-layer perceptron with multiple hidden layers, through combining low-level features to form more abstract high-level representation attributes or features to discover distributed feature representation of data. It can learn the internal rules and representation levels of sample data, and the ultimate goal is to enable machines to have analysis and learning ability like people, and can recognize data such as text, images and sound.

[0097] It should be noted that in the embodiments of the present application, the feature data set of the target object and the action trajectory are involved, when the embodiments of the present application are applied to specific products or technologies, the acquisition, use, processing, etc. of these data need to obtain user permission, authorization or consent, and the collection, use and processing of related data, information, etc. need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0098] The embodiments of the present application provide a network request processing method and device based on artificial intelligence, electronic equipment, computer readable storage medium and computer program product, which can fully and effectively display a conversation scene, and can automatically and accurately predict the interest points to be passed through by the action trajectory.

[0099] The network request processing method based on artificial intelligence provided by the embodiments of the present application can be implemented by the terminal alone, or can be implemented by the terminal and the server in cooperation, for example, the terminal alone bears the network request processing method based on artificial intelligence described below, or the terminal sends a prediction request for a point of interest to the server, the server predicts a target point of interest to be passed through according to the received prediction request for the point of interest, and applies a network service mode matched with the target point of interest to the target object, so as to improve the accuracy of point of interest prediction, and then apply the network service mode to the target point of interest, so as to improve the accuracy of network request processing.

[0100] The electronic device for network request processing provided by the embodiments of the present application can be various types of terminals or servers, wherein the server can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communication, middleware services, domain name services, security services, content distribution networks (CDN), and basic cloud computing services such as big data and artificial intelligence platforms; the terminal can be a smart phone, a tablet computer, a notebook computer, a desktop computer, a smart speaker, a smart watch, a smart television, a vehicle-mounted device, etc., but is not limited thereto. The terminal and the server can be connected directly or indirectly through wired or wireless communication, which is not limited in the present application.

[0101] Taking the server as an example, for example, it can be a server cluster deployed in the cloud, and open artificial intelligence cloud services (AI as a Service, AIaaS) to objects, the AIaaS platform will split several common AI services, and provide independent or packaged services in the cloud. This service mode is similar to an AI theme mall, and all objects can access and use one or more artificial intelligence services provided by the AIaaS platform through the application programming interface.

[0102] For example, one of the artificial intelligence cloud services can be a network request processing service, that is, a server deployed in the cloud encapsulates a network request processing program provided by the embodiments of the present application. The object invokes the network request processing service in the cloud service through the terminal (running a client, such as a WiFi assistant client) to enable the server deployed in the cloud to invoke the encapsulated network request processing program, perform feature extraction processing on the feature data set and the action trajectory of the target object, obtain the object feature of the target object for the action trajectory, perform trajectory classification processing on the action trajectory of the target object based on the object feature of the target object for the action trajectory, obtain the trajectory category to which the action trajectory belongs, perform interest point prediction processing based on meta-learning based on the trajectory category to which the action trajectory belongs, obtain the target interest point to be passed through by the action trajectory, and apply a network service mode matched with the target interest point to the target object, so as to improve the accuracy of interest point prediction, thereby performing targeted network service mode application based on the accurate target interest point, and improving the accuracy of network request processing.

[0103] Referring to Figure 1 , Figure 1 is a schematic diagram of an application scenario of a network request processing system 10 provided by the embodiments of the present application. The terminal 200 is connected to the server 100 through the network 300. The network 300 can be a wide area network or a local area network, or a combination of the two.

[0104] The terminal (running a client, such as a WiFi assistant client) can be used to obtain a prediction request for an interest point. For example, the object carries the terminal to move, changes the current interest point, and the terminal automatically obtains a prediction request for the interest point (including the action trajectory of the target object, such as the object or the terminal).

[0105] In some embodiments, the network request processing plug-in can be implanted in the client running in the terminal 200 to implement the artificial intelligence-based network request processing method locally in the client. For example, the terminal 200 invokes the network request processing plug-in to implement the artificial intelligence-based network request processing method, performs feature extraction processing on the feature data set and the action trajectory of the target object to obtain the object feature of the target object for the action trajectory, performs trajectory classification processing on the action trajectory of the target object based on the object feature of the target object for the action trajectory to obtain the trajectory category to which the action trajectory belongs, performs meta-learning-based interest point prediction processing based on the trajectory category to which the action trajectory belongs to obtain the target interest point to be passed through by the action trajectory, and applies the network service mode matched with the target interest point to the target object to improve the accuracy of interest point prediction, so that the accurate target interest point is used for targeted application of the network service mode, thereby improving the accuracy of network request processing. For example, for the WiFi assistant, when the predicted target interest point is a coffee shop, the WiFi enhancement application is pushed to the target object; when the predicted target interest point is a residential area, the WiFi protection application is pushed to the target object; and when the predicted target interest point is a milk tea shop, the game acceleration application is pushed to the target object.

[0106] In some embodiments, after the terminal 200 obtains the prediction request for the interest point, the network request processing interface of the server 100 (which can be provided in the form of a cloud service, i.e., a network request processing service) is invoked. The server 100 performs feature extraction processing on the feature data set and the action trajectory of the target object based on the prediction request for the interest point to obtain the object feature of the target object for the action trajectory, performs trajectory classification processing on the action trajectory of the target object based on the object feature of the target object for the action trajectory to obtain the trajectory category to which the action trajectory belongs, performs meta-learning-based interest point prediction processing based on the trajectory category to which the action trajectory belongs to obtain the target interest point to be passed through by the action trajectory, and sends the target interest point to be passed through by the action trajectory to the terminal 200. The terminal 200 applies the network service mode matched with the target interest point to the target object to improve the accuracy of interest point prediction, so that the accurate target interest point is used for targeted application of the network service mode, thereby improving the accuracy of network request processing. For example, for the WiFi assistant, when the predicted target interest point is a coffee shop, the WiFi enhancement application is pushed to the target object; when the predicted target interest point is a residential area, the WiFi protection application is pushed to the target object; and when the predicted target interest point is a milk tea shop, the game acceleration application is pushed to the target object.

[0107] In some embodiments, the terminal or the server can implement the artificial intelligence-based network request processing method provided in the embodiments of the present application by running a computer program. The computer program is, for example, Figure 1The client running in the terminal 200 shown in the figure, for example, a computer program, can be a native program or a software module in an operating system; can be a native application (APP) that needs to be installed in the operating system to run; can also be a mini-program that only needs to be downloaded into a browser environment to run; and can also be a mini-program that can be embedded into any APP. In summary, the above computer program can be any form of application program, module or plug-in.

[0108] In some embodiments, a plurality of servers can be composed into a blockchain, and the server 100 is a node on the blockchain, and there can be an information connection between each node in the blockchain, and the nodes can transmit information through the above information connection. Among them, the data related to the network request processing method based on artificial intelligence provided by the embodiments of the application (for example, the logic of network request processing, the target interest point to be passed through by the action track) can be saved on the blockchain.

[0109] The structure of the electronic device for network request processing provided by the embodiments of the application is described below. Referring to Figure 2 , Figure 2 is a structural schematic diagram of the electronic device 500 for network request processing provided by the embodiments of the application. Taking the electronic device 500 as a terminal as an example, Figure 2 The electronic device 500 for knowledge generation shown in the figure includes at least one processor 510, a memory 550, at least one network interface 520 and a user interface 530. Each component in the electronic device 500 is coupled together through a bus system 540. It can be understood that the bus system 540 is used to realize the connection communication between the components. In addition to including a data bus, the bus system 540 also includes a power bus, a control bus and a status signal bus. However, for the purpose of clear illustration, all kinds of buses are marked as the bus system 540 in Figure 2 .

[0110] The processor 510 can be an integrated circuit chip with signal processing capability, such as a general-purpose processor, a digital signal processor (DSP), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc., wherein the general-purpose processor can be a microprocessor or any conventional processor.

[0111] The memory 550 includes volatile memory or nonvolatile memory, and can include both volatile and nonvolatile memory. Among them, the nonvolatile memory can be a read-only memory (ROM), and the volatile memory can be a random access memory (RAM). The memory 550 described in the embodiments of the present application is intended to include any suitable type of memory. The memory 550 can optionally include one or more storage devices that are physically separate from the processor 510.

[0112] In some embodiments, the memory 550 can store data to support various operations, examples of which include programs, modules, and data structures or subsets or supersets thereof, which are exemplarily illustrated below.

[0113] The operating system 551 includes system programs for processing various basic system services and performing hardware-related tasks, such as a framework layer, a core library layer, a driver layer, etc., for implementing various basic services and processing hardware-based tasks.

[0114] The network communication module 552 is used to reach other computing devices via one or more (wired or wireless) network interfaces 520, exemplary network interfaces 520 including Bluetooth, wireless compatibility authentication (WiFi), and universal serial bus (USB), etc.

[0115] In some embodiments, the artificial intelligence-based network request processing apparatus provided by the embodiments of the present application can be implemented in a software manner, and the artificial intelligence-based network request processing apparatus provided by the embodiments of the present application can be provided in various software embodiments, including various forms such as application programs, software, software modules, scripts, or codes.

[0116] Figure 2 An artificial intelligence-based network request processing apparatus 555 stored in the memory 550 is shown, which can be software in the form of programs and plug-ins, and includes a series of modules, including an acquisition module 5551, a feature extraction module 5552, a classification module 5553, a prediction module 5554, and an application module 5555. These modules are logical, and thus can be combined or further split according to the functions implemented, and the functions of each module will be described below.

[0117] As described previously, the artificial intelligence-based network request processing method provided by the embodiments of the present application can be implemented by various types of electronic devices. Referring to Figure 3A , Figure 3A is a flowchart of the artificial intelligence-based network request processing method provided by the embodiments of the present application, in combination with Figure 3AThe steps shown are described.

[0118] In step 101, a feature data set of a target object and a behavior trajectory are acquired, wherein the behavior trajectory includes at least one historical interest point of a route.

[0119] Before acquiring the feature data set of the target object and the behavior trajectory, a permission pop-up window for information acquisition is displayed on a display interface of a terminal carried by the target object, the permission pop-up window is used to prompt that the acquisition, use, processing, etc. of the data of the feature data set and the behavior trajectory need to obtain user permission, authorization or consent, when a consent button on the permission pop-up window is triggered, it indicates that the acquisition, use, processing, etc. of the data of the feature data set and the behavior trajectory of the target object has obtained user permission, authorization or consent.

[0120] As an example of acquiring the feature data set of the target object and the behavior trajectory, in the case that the acquisition, use, processing, etc. of the data of the feature data set and the behavior trajectory of the target object has obtained user permission, authorization or consent, when the target object (i.e. a real object) carries a terminal to move and changes the current interest point, the terminal automatically acquires a prediction request (including the behavior trajectory of the target object) for the interest point and sends the prediction request for the interest point to a server, the server analyzes the prediction request for the interest point, acquires an account of the target object, and acquires the feature data set of the target object in a distributed file system (HDFS, The Hadoop Distributed File System) based on the account of the target object.

[0121] It should be noted that the feature data set includes a basic feature data set and a business vertical type characteristic. Based on the object historical behavior data, rich basic features are constructed, including object basic attributes, device basic attributes, network connection attributes, etc. For example, object basic attributes (age: male), device basic attributes (mobile phone brand: XX), network connection attributes (Wi-Fi connection times this week is 10 times). Based on the business characteristics, business vertical type features are constructed, and the vertical type features include the click rate and conversion rate of the object to specific type recommendation information. Before acquiring the feature data set of the target object, various feature data sets of objects can be constructed in advance and stored in the HDFS, so that the feature data set of the target object can be quickly acquired according to the account of the target object subsequently.

[0122] It should be noted that in the embodiments of the present application, the feature data set of the target object and the behavior trajectory are involved, when the embodiments of the present application are applied to specific products or technologies, the acquisition, use, processing, etc. of these data need to obtain user permission, authorization or consent, and the collection, use and processing of related data, information, etc. need to comply with relevant laws, regulations and standards of relevant countries and regions.

[0123] In step 102, feature extraction processing is performed on the feature data set and movement trajectory of the target object to obtain the object features of the target object in relation to the movement trajectory.

[0124] For example, after obtaining the feature data set and movement trajectory of the target object, feature extraction needs to be performed using a feature extraction model to obtain the object features of the target object related to the movement trajectory. Subsequently, trajectory classification can be performed based on the object features of the target object related to the movement trajectory. The object features of the target object related to the movement trajectory include the feature data set and features that accurately represent the target object and its movement trajectory.

[0125] See Figure 3B , Figure 3B This is an optional flowchart illustrating an artificial intelligence-based network request processing method provided in an embodiment of this application. Figure 3A Step 102 shown is achieved through Figure 3B Steps 1021-1022 are implemented as follows: In step 1021, the feature data set of the target object and the movement trajectory are fused to obtain the fused information of the target object; the fused information of the target object is subjected to feature extraction to obtain the object features of the target object in relation to the movement trajectory.

[0126] For example, the feature data set and movement trajectory of the target object can be used as input for feature extraction to extract the object features of the target object related to the movement trajectory. The fusion of the feature data set and movement trajectory of the target object can be achieved through the following schemes: Scheme 1: Concatenate the feature data set and movement trajectory of the target object to obtain fused information. This simple concatenation operation allows input of all the feature data set and movement trajectory of the target object, avoiding omissions. Scheme 2: Perform a weighted summation of the feature data set and movement trajectory of the target object to obtain fused information. This takes into account the importance of the feature data set and movement trajectory of the target object, accurately extracting the object features of the target object related to the movement trajectory.

[0127] In some embodiments, the feature extraction processing is implemented by a feature extraction model, and the feature extraction model includes a plurality of cascaded feature extraction layers; the feature extraction processing is performed on the fusion information of the target object to obtain the object feature of the target object for the action trajectory, including: performing feature extraction processing on the fusion information of the target object by a first feature extraction layer in the plurality of cascaded feature extraction layers; outputting the feature extraction result of the first feature extraction layer to a subsequent cascaded feature extraction layer, and continuing to perform feature extraction processing and feature extraction result output by the subsequent cascaded feature extraction layer until the last feature extraction layer; and taking the feature extraction result output by the last feature extraction layer as the object feature of the target object for the action trajectory.

[0128] For example, the feature extraction model (implemented by the feature extractor) includes cascaded feature extraction layers, and the feature extraction is sequentially performed by the cascaded feature extraction layers, so that the object feature with progressive precision can be obtained through the cascaded feature extraction processing, and finally the object feature accurately containing all semantic information is obtained to accurately represent the feature data set of the target object and the action trajectory.

[0129] For example, the feature extraction processing is performed on the fusion information of the target object by the first feature extraction layer; the feature extraction result of the first feature extraction layer is output to the second feature extraction layer, and the feature extraction processing and the feature extraction result output are continued by the second feature extraction layer, and the feature extraction processing and the feature extraction result output are iterated, and the feature extraction result output by the last feature extraction layer is taken as the object feature of the target object for the action trajectory.

[0130] In some embodiments, the feature extraction processing and the feature extraction result output are continued by the subsequent cascaded feature extraction layer, including: performing the following processing by the jth feature extraction layer in the plurality of cascaded feature extraction layers: performing random convolution processing on the feature extraction result of the (j-1)th feature extraction layer to obtain the jth deep layer feature; performing pooling processing on the jth deep layer feature to obtain the feature extraction result of the jth feature extraction layer; outputting the feature extraction result of the jth feature extraction layer by the jth feature extraction layer; wherein j is an increasing natural number and the value range is 1

[0131] For example, as shown in FIG. 2, the feature extraction model includes a plurality of cascaded feature extraction layers, and the feature extraction processing is performed on the fusion information of the target object by the plurality of cascaded feature extraction layers to obtain the object feature of the target object for the action trajectory. Figure 9As shown, each feature extraction layer includes a random convolutional layer and a pooling layer. The feature extraction result of the (j-1)th feature extraction layer is processed by the random convolutional layer in the j-th feature extraction layer. The convolutional layer is used to mine deep features from the input sequence (the feature data set of the target object and its movement trajectory), obtaining the j-th deep feature. The j-th deep feature is then pooled by the pooling layer in the j-th feature extraction layer to remove redundant features, resulting in the feature extraction result of the j-th feature extraction layer. Finally, the feature extraction result of the j-th feature extraction layer is output.

[0132] It's important to note that random convolutional layers are similar to convolutional layers in convolutional neural networks (CNNs), learning local feature representations of the input vector through convolutional operations. However, unlike convolutional layers in CNNs, each element in the convolutional kernel of a random convolutional layer is randomly generated, eliminating the need for fine-tuning and thus improving model training speed.

[0133] In step 103, based on the object characteristics of the target object's movement trajectory, the movement trajectory of the target object is classified to obtain the trajectory category to which the movement trajectory belongs.

[0134] For example, after extracting the object features of the target object's movement trajectory, a trajectory classification model is used to classify the trajectory to obtain the trajectory category to which the movement trajectory belongs. This allows for subsequent prediction of points of interest based on the trajectory category to which the movement trajectory belongs, thereby improving the accuracy of point of interest prediction.

[0135] For example, the trajectory category is the object pattern to which the action trajectory belongs, such as the trajectory category of working type (i.e., office worker), school type (i.e., student), shopping type (i.e., shopper), game type (game player), etc.

[0136] See Figure 3C , Figure 3C This is an optional flowchart illustrating an artificial intelligence-based network request processing method provided in an embodiment of this application. Figure 3A Step 102 shown is achieved through Figure 3C Steps 1031-1032 are implemented as follows: In step 1031, the object features of the target object's movement trajectory are subjected to temporal encoding processing to obtain the temporal features of the movement trajectory; in step 1032, the temporal features of the movement trajectory are mapped to obtain the trajectory category to which the movement trajectory belongs.

[0137] For example, the trajectory classification model is implemented by a bidirectional long short-term memory network (BLSTM), which encodes the object features of the target object for the action trajectory in time sequence by the BLSTM, extracts the past and future time information of the object features of the target object for the action trajectory, obtains the time sequence features of the action trajectory, and maps the time sequence features of the action trajectory to obtain the trajectory category to which the action trajectory belongs.

[0138] In some embodiments, the object features of the target object for the action trajectory are encoded in time sequence to obtain the time sequence features of the action trajectory, including: performing forward time sequence encoding processing on the object features of the target object for the action trajectory to obtain forward time sequence features of the action trajectory; performing reverse time sequence encoding processing on the object features of the target object for the action trajectory to obtain reverse time sequence features of the action trajectory; and performing splicing processing on the forward time sequence features and the reverse time sequence features to obtain the time sequence features of the action trajectory.

[0139] For example, the BLSTM is a combination of a forward time recurrent neural network algorithm (LSTM) and a reverse LSTM algorithm, which extracts the forward time sequence features of the action trajectory by the forward LSTM, extracts the reverse time sequence features of the action trajectory by the reverse LSTM, splices the forward time sequence features and the reverse time sequence features to obtain the time sequence features of the action trajectory, and reflects the time dependence of the action trajectory of the target object through the time sequence features of the action trajectory, which is used to extract the past and future time information of the action trajectory of the target object.

[0140] In some embodiments, the object features of the target object for the action trajectory are encoded in time sequence to obtain the time sequence features of the action trajectory, including: performing forward time sequence encoding processing on the object features of the target object for the action trajectory to obtain forward time sequence features of the action trajectory; performing reverse time sequence encoding processing on the object features of the target object for the action trajectory to obtain reverse time sequence features of the action trajectory; and performing splicing processing on the forward time sequence features and the reverse time sequence features to obtain the time sequence features of the action trajectory.

[0141] For example, the forward LSTM is composed of a series of memory units. The memory unit contains three gate structures: the forget gate, the input gate and the output gate. The forward LSTM can determine the retention and discard of input information through the three gate structures, and realize the cyclic update of input information. The input sequence xt (the input of the (i-1)th forward time sequence feature at the tth moment) passes through the forget gate to determine the retention and forgetting of information. The retention and forgetting of information is determined by a sigmoid function. If the sigmoid value is 0, all information is discarded; if it is 1, all information is retained.

[0142] Based on the (i-1)th forward time sequence feature, the object feature is subjected to forward screening processing based on the forget gate to obtain the forward output feature of the forget gate , wherein, represents the output of the forget gate at the tth moment, represents a sigmoid function, represents the weight of the forget gate, represents a bias.

[0143] Based on the (i-1)th forward time sequence feature, the object feature is subjected to forward update processing based on the input gate to obtain the forward output feature of the input gate and , wherein, represents the output at the tth moment, which can determine the updated information. represents the candidate cell information.

[0144] Based on the (i-1)th forward time sequence feature, the object feature is subjected to forward mapping processing based on the output gate to obtain the forward output feature of the output gate The output gate determines the output information, and determines which information to output through the sigmoid function.

[0145] The forward output feature of the forget gate, the forward output feature of the input gate and the forward output feature of the output gate are subjected to nonlinear mapping processing to obtain the ith forward time sequence feature .

[0146] In some embodiments, the object feature of the target object is reversely time-coded for the action trajectory to obtain a reverse time feature of the action trajectory, including: performing a reverse screening processing of the object feature based on a forgetting gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the forgetting gate; performing a reverse update processing of the object feature based on an input gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the input gate; performing a reverse mapping processing of the object feature based on an output gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the output gate; performing a nonlinear mapping processing on the reverse output feature of the forgetting gate, the reverse output feature of the input gate and the reverse output feature of the output gate to obtain the e th reverse time feature; taking a set of M reverse time features as the reverse time feature of the action trajectory; wherein e is a natural number decreasing and the value range is 1≤e≤M-1, M is the number of features of the object feature, and M is a positive integer greater than 1.

[0147] For example, the reverse LSTM is composed of a series of memory cells. The memory cell contains three gate structures: forgetting gate, input gate and output gate. The reverse LSTM can determine the retention and discard of input information through the three gate structures to realize the cyclic update of input information. The input sequence xt (the input at the t th moment (the (i-1)th reverse time feature)) passes through the forgetting gate to determine the retention and forgetting of information. The retention and forgetting of information is determined by a sigmoid function. If the sigmoid value is 0, all information is discarded; if it is 1, all information is retained.

[0148] performing a reverse screening processing of the object feature based on a forgetting gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the forgetting gate , wherein, represents the output of the forgetting gate at t moment, represents a sigmoid function, represents the weight of the forgetting gate, represents the bias.

[0149] performing a reverse update processing of the object feature based on an input gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the input gate and , wherein, represents the output at t moment, which can determine the updated information. represents the candidate cell information.

[0150] performing a reverse mapping processing of the object feature based on an output gate based on the (e+1)th reverse time feature to obtain a reverse output feature of the output gate The output gate determines the output information, which is determined by a sigmoid function to output which information.

[0151] The reverse output feature of the forget gate, the reverse output feature of the input gate, and the reverse output feature of the output gate are subjected to nonlinear mapping processing to obtain an i-th reverse time sequence feature .

[0152] In step 104, meta-learning-based point of interest prediction processing is performed based on the trajectory category to which the action trajectory belongs, to obtain a target point of interest that the action trajectory will pass through.

[0153] For example, the meta-learning model is used to perform meta-learning-based point of interest prediction processing on the trajectory category to which the action trajectory belongs, to obtain a target point of interest that the action trajectory will pass through (i.e., the next point of interest that the action trajectory will pass through). By using the meta-learning model to predict the point of interest, the sparsity problem of the action trajectory can be alleviated, and the model is forced to be insensitive to negative samples of the action trajectory, thereby improving the accuracy of the point of interest prediction.

[0154] In some embodiments, the point of interest prediction processing is implemented by a meta-learning model, and the training process of the meta-learning model includes: performing meta-learning-based point of interest prediction processing based on the trajectory category to which the action trajectory belongs, to obtain a target point of interest that the action trajectory will pass through, including: constructing a query set and a support set based on a plurality of action trajectory samples, the support set including at least one support subset, each support subset including at least one first action trajectory sample in the plurality of action trajectory samples, different support subsets including action trajectory samples of different trajectory categories, the query set including at least one second action trajectory sample in the plurality of action trajectory samples; performing the following processing by using an initialized meta-learning model: performing meta-learning-based point of interest prediction processing on the first action trajectory samples included in the support subset to obtain first predicted points of interest that the first action trajectory samples will pass through; performing meta-learning-based point of interest prediction processing on the second action trajectory samples included in the query set to obtain second predicted points of interest that the second action trajectory samples will pass through; constructing a loss function of the meta-learning model based on the first predicted points of interest, the point of interest labels corresponding to the first action trajectory samples, the second predicted points of interest, and the point of interest labels corresponding to the second action trajectory samples; and updating the parameters of the meta-learning model based on the loss function, and using the updated parameters of the meta-learning model as the parameters of the trained meta-learning model.

[0155] For example, after determining the value of the loss function of the meta-learning model, it can be determined whether the value of the loss function reaches a preset threshold. When the value of the loss function does not reach the preset threshold, an error signal of the meta-learning model is determined based on the loss function, the error signal is back-propagated in the meta-learning model, and the model parameters of each layer are updated during the propagation.

[0156] Here, the back propagation is explained, the training sample data is input to the input layer of the neural network model, through the hidden layer, finally reaches the output layer and outputs the result, which is the forward propagation process of the neural network model. Since there is an error between the output result of the neural network model and the actual result, the error between the output result and the actual value is calculated, and the error is propagated from the output layer to the hidden layer until it propagates to the input layer. In the process of back propagation, the value of the model parameter is adjusted according to the error, that is, the loss function is constructed according to the error between the output result and the actual value, and the partial derivative of the loss function with respect to the model parameter is calculated layer by layer to generate the gradient of the loss function with respect to each layer model parameter. Since the direction of the gradient indicates the direction of error expansion, the gradient of the model parameter is taken as the opposite, and the sum of the original parameters of each layer model is taken as the updated parameters of each layer model, so as to reduce the error caused by the model parameters. The above process is iterated until convergence. Among them, the meta-learning model belongs to the neural network model.

[0157] In step 105, the network service mode matched with the target point of interest is applied to the target object.

[0158] For example, after obtaining the next point of interest to be passed through the action trajectory, the network service mode application can be applied to the next point of interest, thereby improving the accuracy of network request processing. For example, when the next point of interest is a coffee shop or the like, WiFi enhancement application is applied, that is, the speed of WiFi is improved; when the next point of interest is a family, a residential area or a hotel or the like, WiFi protection application is applied, that is, WiFi protection function is realized; when the next point of interest is an Internet bar or the like, game acceleration application is applied, that is, the network speed of the game is accelerated to avoid lag.

[0159] In some embodiments, the network service mode matched with the target point of interest is applied to the target object, including: based on the target point of interest, querying the correspondence between different points of interest and candidate network service modes, and taking the queried candidate network service mode as the network service mode matched with the target point of interest; displaying the recommendation information of the network service mode; in response to the application operation of the network service mode based on the recommendation information, applying the network service mode to the terminal corresponding to the target object.

[0160] For example, different points of interest may be associated with different network service modes. After determining the network service mode matched with the target point of interest, the network service mode matched with the target point of interest can be recommended to the target object. In response to the manual selection of the target object, the network service mode matched with the target point of interest can be applied. Thus, the actual situation of the current target object can be considered, and whether the network service mode needs to be switched can be manually selected.

[0161] In some embodiments, the network service mode matched with the target point of interest is applied to the target object, including: querying the correspondence between different points of interest and candidate network service modes based on the target point of interest, and taking the queried candidate network service mode as the network service mode matched with the target point of interest; and applying the network service mode matched with the target point of interest to the terminal corresponding to the target object.

[0162] For example, different points of interest can be associated with different network service modes, and after determining the network service mode matched with the target point of interest, the network service mode can be automatically switched to realize intelligent application of the network service mode.

[0163] It should be noted that the network service mode matched with the target point of interest is applied to the terminal corresponding to the target object, including: obtaining the distance from the terminal corresponding to the target object to the target point of interest; and when the distance is less than a distance threshold, the network service mode matched with the target point of interest is applied to the terminal corresponding to the target object.

[0164] For example, only when the target point of interest is approached, the network service mode is automatically switched to avoid subsequent changes in the travel of the target object, so that the target object does not reach the target point of interest, and the wrong network service mode is applied.

[0165] It should be noted that when the network service mode matched with the target point of interest is multiple, the network service mode matched with the target point of interest is applied to the terminal corresponding to the target object, including: obtaining a target time period when the target point of interest is reached; determining the network service mode matched with the target time period from the multiple network service modes matched with the target point of interest; and applying the network service mode matched with the target time period to the terminal corresponding to the target object.

[0166] For example, considering that different time periods can apply different network service modes, when the network service mode matched with the target point of interest is multiple, the network service mode matched with the current time is determined, and the network service mode matched with the current time is applied to the terminal corresponding to the target object, thereby improving the accuracy of the application of the network service mode.

[0167] In the following, an exemplary application of the embodiments of the present application in an actual application scenario will be described.

[0168] The embodiments of the present application can be applied to various object movement mode identification scenarios, for example, for a WiFi assistant, when the next POI is predicted, the WiFi application in various scenarios, such as WiFi enhanced package, is timely pushed.

[0169] It should be noted that WiFi Assistant is a WiFi management software that supports hundreds of millions of public WiFi hotspots. It allows one-click connection without entering a password and comprehensively evaluates these hotspots based on WiFi standards, including security, connection speed, and network quality, to ensure that there are no zombie, risky, or fake WiFi networks.

[0170] An object's WiFi usage needs are often closely related to its movement patterns. For example, if an object's current POI is in a coffee shop, it may have a strong need for network speed. In this case, pushing a "WiFi enhancement package" based on the product can provide a better user experience. Therefore, building a highly accurate object movement pattern recognition system is of great significance for product operations.

[0171] The mapping between WiFi Assistant's various scenarios and strongly matched POIs is shown below:

[0172] Scenario 1, such as Figure 4 The WiFi enhancement package shown is designed for scenarios with high network requirements. In such scenarios, users have high demands for network speed and stability. The WiFi Assistant's enhancement package can technically improve WiFi stability and speed. When the motion pattern recognition model identifies that the user is in a coffee shop or internet cafe, it will proactively push features such as... Figure 4 The WiFi enhancement package shown is available to users. Clicking the "Experience Now" button (401) will improve WiFi speed. Data shows that the percentage of users clicking to use the WiFi enhancement package at POIs (Coffee Shops and Internet Cafes) is 192% higher than that of randomly selected users.

[0173] Scenario 2, such as Figure 5 The home WiFi protection plan shown (i.e., WiFi protection function) is used by the user to fill in their home WiFi information, so that the product can provide WiFi protection functions. These protection functions include: network speed maintenance, management of the number of connected WiFi devices, and prevention of unauthorized access. When the mobile pattern recognition model identifies that the user will be in a home, residential community, or hotel, it will proactively push information such as... Figure 5 The home WiFi protection plan shown is available to the target user. Clicking the "Confirm" button (501) will activate the WiFi protection function. Data shows that the percentage of users who clicked to join the home WiFi protection plan under POIs (Homes, Residential Communities, and Hotels) was 410% higher than that of randomly selected users in the general survey.

[0174] Scenario 3, Application acceleration (e.g., ... Figure 6A The video acceleration shown and as Figure 6BThe game acceleration shown) is a WiFi assistant based on technical means to provide single-point application (App) network speed protection and acceleration service functions of dedicated line network. When the object is identified to be in a scenario such as a milk tea shop through a mobile mode recognition model, the application acceleration function is actively pushed to the object. For example, when the object is watching a video, the object is pushed to, for example Figure 6A The video acceleration shown, click the "start acceleration" button 601, to accelerate the network speed of the video, to avoid lag; when the object is playing a game, the object is pushed to, for example Figure 6B The game acceleration shown, click "start acceleration" 602, to accelerate the network speed of the game, to avoid lag. Data shows that the proportion of objects that click to use application acceleration in POIs such as milk tea shops is 704% higher than that of random objects in the market.

[0175] The object mobile mode recognition scenario is described below:

[0176] The related art has the following object mobile mode recognition method:

[0177] A) Based on artificial experience to determine data rules. Product operation is based on business experience to set artificial recognition rules, such as "age 15-20, last visited POI is game room", the object is more likely to be a bus or subway next time. However, this solution not only has a very limited number of rules, but also cannot capture the high-dimensional feature information of the interaction between rules, and most importantly, cannot determine the optimal parameters of each rule.

[0178] B) Based on non-deep learning data mining method, through the method of constructing multi-dimensional features and model training, to predict the probability of the current object belonging to different mobile modes. However, in the object mobile mode recognition scenario, due to the complexity of the behavior characteristics, the feature information is difficult to express explicitly in data representation.

[0179] To address the aforementioned issues, this application provides an AI-based network request processing method, successfully applied in object movement pattern recognition scenarios. To improve the performance of location recovery and trajectory user link recognition in movement pattern recognition, this application utilizes an algorithmic framework combining a feature extractor (implemented by an Extreme Learning Machine) and a Bidirectional Long Short-Term Memory (BLSTM) network. The Extreme Learning Machine with a local receptive field is used to mine deep features of the input sequence, and pooling layers are used to remove redundant information. The BLSTM network is used to extract past and future time information of the input sequence to effectively reflect the sequence's dependence on time. To address the issues of data sparsity and model susceptibility to negative samples, this application proposes a meta-learning framework (MetaMove). Based on a meta-learning network, unlabeled trajectory data is used at both the meta-learning and adaptive levels to alleviate data sparsity and force the model to be insensitive to negative samples. This application can improve the accuracy of object movement pattern recognition.

[0180] like Figure 7 The following is a detailed description of the AI-based network request processing method provided in the embodiments of this application, as shown in the architecture diagram:

[0181] The technical solution architecture of this application embodiment includes the following parts:

[0182] 1) Sample preparation. Based on manual annotation and business experience, obtain positive and negative training samples that are strongly relevant to the business, have normal data distribution, and have a reasonable set of target object feature data.

[0183] 2) Feature Construction. Construct basic features for the training samples, and based on the vertical properties of the features, combine the time dimension and different feature processing methods to produce high-dimensional feature vectors.

[0184] 3) Feature Extractor. This algorithm is used to extract local spatial features from a sequence and fully consider long-term dependencies within the sequence.

[0185] 4) Meta-learning model (MetaMove model). This model is based on meta-learning to mine human movement patterns and unifies movement prediction and classification models within a single framework to output the probability of predicting the next POI of an object.

[0186] The following is combined with Figure 8 Detailed explanation of sample preparation, feature construction, feature extractor, BLSTM, and MetaMove model:

[0187] Sample preparation can be carried out in the following ways:

[0188] Step 1, based on artificial labeling or business logic, obtain seed objects with label markers.

[0189] For example, based on screening rules, a batch of seed objects are recalled, then filtered based on artificial screening, and finally verified based on business logic.

[0190] Step 2, obtain the basic features of seed objects.

[0191] For example, basic features include some non-private behavior data of objects in the application, such as whether to install mobile manager, whether to use mobile manager harassment interception function, etc.

[0192] Step 3, calculate the evaluation index of abnormal object type.

[0193] For example, in real business scenarios, there may be false objects and computer-controlled mobile phones. In order to eliminate the influence of non-real objects on modeling analysis, abnormal object detection indicators are set based on business experience, such as object traffic usage in the application, time distribution of traffic generation, etc.

[0194] Step 4, filter abnormal seed objects based on distribution anomaly theorem.

[0195] For example, use the "Pareto criterion" to judge the standard of abnormal value. The specific method is as follows: assuming that a set of detection data only contains random errors, the standard deviation is obtained by calculating and processing, and an interval is determined according to the set probability, then any error exceeding this interval is considered not to be random error but gross error, and the data containing the gross error should be removed.

[0196] Feature construction is achieved in the following way:

[0197] Step 5, construct object features.

[0198] It should be noted that object features include basic features and vertical type features, and the construction process of basic features and vertical type features is as follows:

[0199] 1. Construct basic features. Based on object historical behavior data, a rich set of target object feature data is constructed, including: object basic attributes, device basic attributes, network connection attributes, etc. For example, object basic attributes (age: male), device basic attributes (mobile phone brand: XX), network connection attributes (Wi-Fi connection times this week is 10 times).

[0200] 2. Based on business characteristics, construct vertical type features. Vertical type features include object click rate and conversion rate of specific type recommendation information.

[0201] Step 6, construct aggregated features combined with time dimension.

[0202] For example, the basic features and business features of different time spans are aggregated to obtain aggregated features. For example, the aggregated features of the object in the last half year, the last three months, the last month, and the last week are calculated, and the aggregation methods are selected from three kinds of summation, median, average (average pooling), and standard deviation. As shown in the following table, the average pooling is used to aggregate a basic feature to obtain an aggregated feature. Figure 9

[0203] Step 7, feature processing: normalizing numerical features and discretizing non-numerical features.

[0204] For example, the discretization processing includes the following methods:

[0205] i. One-Hot Encoding. For example, for the object gender feature, One-Hot Encoding becomes: male: (1, 0), female: (0, 1).

[0206] ii. Count Encoding. For example, for the POI feature of the object, Count Encoding is used to identify the interest degree of the object and the POI. For example, the object went to the POI “Food-Chinese food-Cantonese food” 3 times in the week.

[0207] iii. Consolidation Encoding. There are multiple values under certain category variables, which can be summarized into the same information. For example, the multiple values of the Android phone system version feature include “4.2”, “4.4” and “5.0”, which can be summarized as “low version Android system” based on experience. Consolidation Encoding processing can bring greater positive benefits than directly One-Hot Encoding the “Android system version” feature.

[0208] Step 8, merging the processed features and storing them offline in a distributed file system (HDFS, The Hadoop Distributed File System).

[0209] For example, after merging the processed features and storing them offline in HDFS, it is convenient for subsequent quick access of features, that is, based on the identification of the target object, the object features (i.e., the feature data set of the target object) can be quickly obtained from HDFS. For each object, the data input to the model is an N*1 numerical vector, such as (1, 0, 31, 4, 0.2, 9.3, 8.8, …, 0, 0, 1, 2, 34).

[0210] ​Step 9, input the input sequence (i.e. object features) into the feature extractor, use multiple convolution layers to mine deep features of the input sequence, use a pooling layer to remove redundant features, and use the output of the last pooling layer as the input of the BLSTM layer.

[0211] It should be noted that, as shown in Figure 10 The feature extractor and the BLSTM algorithm framework include three parts, namely the feature extractor, the BLSTM layer and the Softmax output layer. The three parts included in the feature extractor and the BLSTM algorithm framework will be described below:

[0212] Regarding the feature extractor:

[0213] The feature extractor includes three random convolution layers and pooling layers which are alternately arranged. The multiple convolution layers are used to mine deep features of the input sequence (input features), and the pooling layer is used to remove redundant features. The random convolution layer and the pooling layer are alternately arranged, and the output of the last pooling layer is used as the input of the BLSTM layer.

[0214] It should be noted that the random convolution layer is similar to the convolution layer in the convolution neural network (CNN, Convolutional Neural Network), which learns the local feature representation of the input vector through convolution operation. Unlike the convolution layer in the CNN, each element in the convolution kernel in the random convolution layer is randomly generated and does not need to be fine-tuned. Assuming that the input of the model is a vector with a size of d*1, the convolution kernel is r*1, and there are K feature maps, according to the principle of convolution operation, the size of each feature map is d-r+1, and the output value of each random convolution layer is calculated in the manner shown in formula (1):

[0215]

[0216] wherein, represents the output value at the node (i, 1) in the feature map.

[0217] As shown in the one-dimensional convolution process of the random convolution layer shown in Figure 11 , after the sequence (1, 2, 3) is convolved by the convolution kernel, the sequence (1, 2) is obtained.

[0218] It should be noted that the pooling layer of the embodiment of the application selects the maximum pooling, which can remove redundant information, reduce the feature dimension, and further reduce the algorithm parameters and complexity. Assuming that the pooling size is s, the pooling layer will not change the number of feature maps, and the output of the kth feature map is calculated in the manner shown in formula (2):

[0219]

[0220] wherein,​

[0221] like Figure 12 The one-dimensional max pooling process shown obtains the sequence (2,4) after max pooling the sequence (1,2,3,4).

[0222] Step 10: By combining forward LSTM and backward LSTM algorithms, the time dependency of the input sequence is reflected, and the past and future time information of the input sequence is extracted.

[0223] It should be noted that the BLSTM layer is a combination of the forward temporal recurrent neural network algorithm (LSTM, Long Short-Term Memory) and the backward LSTM algorithm. It can reflect the temporal dependencies of the input sequence and is used to extract past and future temporal information from the input sequence. Assuming the hidden layer outputs of the forward LSTM and backward LSTM are respectively... and ,Will and By concatenating the components, we obtain the hidden layer output of the BLSTM. .

[0224] like Figure 13 The BLSTM algorithm structure shown has two LSTMs in opposite directions connected to the input xt at each time step. The output yt at the current time step is a combination of the two LSTMs, that is, the forward and backward LSTMs jointly determine the output at the current time step.

[0225] It should be noted that LSTM is composed of a series of memory units. Each memory unit contains three gate structures: a forget gate, an input gate, and an output gate. LSTM can use these three gate structures to determine the retention and discarding of input information, thereby achieving cyclical updates of the input information.

[0226] The input sequence xt (the input at time t) passes through a forget gate to determine whether to retain or forget information. The retention or forgetting of information is determined by a sigmoid function. If the sigmoid value is 0, all information is discarded; if it is 1, all information is retained. The calculation formula is shown in formula (3):

[0227]

[0228] in, It is the input at time t. This represents the output of the forget gate at time t. This represents the sigmoid function. Indicates the weight of the forget gate. Indicates bias.

[0229] The calculation formula of the input gate is shown in equations (4) and (5):

[0230]

[0231]

[0232] wherein, represents the output at time t, and the updated information can be determined. represents the candidate cell information.

[0233] The output gate determines the output information, and determines which information to output through a sigmoid function. The calculation formula is shown in equation (6):

[0234]

[0235] Cell state The Hadamard product of the output after the tanh function and the output of the output gate is the output of the hidden layer at time t , and the calculation formula is shown in equations (7) and (8):

[0236]

[0237]

[0238] wherein, represents the Hadamard product.

[0239] In summary, the calculation formula of the BLSTM is shown in equations (9) to (11):

[0240]

[0241]

[0242]

[0243] Step 11: Classify the input sequence using an activation function to obtain the category to which the trajectory belongs.

[0244] For example, classify the input sequence using an activation function (such as a softmax function). The softmax function is defined as shown in equation (12):

[0245]

[0246] wherein, N represents the number of input data, j represents the category, represents the i-th input, w represents a learning parameter, represents is the probability of category j.

[0247] Step 12: Predict the next POI of the object based on the MetaMove model.

[0248] For example, before predicting the next POI of an object using the MetaMove model, the MetaMove model needs to be trained, and the training process is as follows:

[0249] like Figure 14 The MetaMove model shown calculates the loss function using the support set and query set, updates the model parameters based on the loss function, and finally uses the trained model to predict POIs.

[0250] It should be noted that task sampling is required before model training.

[0251] In the meta-learning environment, the dataset is divided into a meta-training set D_train and a meta-test dataset D_test. The model M trained on D_train is transferred to D_test for testing. The class labels in D_test are not visible in D_train. This embodiment changes the testing paradigm during meta-training, allowing the output of Points of Interest (POIs) not present in the training set.

[0252] During model training, a predefined loss function L is used to reduce the loss value, thereby reducing prediction error. The MetaMove model accelerates model learning by generating various sampling datasets, especially for unseen class labels. The detailed process is shown in the pseudocode below:

[0253] Input: Training dataset The number of meta-tasks is k, and the parameters are...

[0254] Output: Meta-task set

[0255] When i = 1, 2, ..., k, the following processing is performed:

[0256] Take from object set U A collection of objects

[0257] when exist When in transit, perform the following processing:

[0258] from trajectory dataset The λ trajectories sampled from the middle are placed into the support set.

[0259] from trajectory dataset sample η trajectories (different from the ones already taken out above) into the query set

[0260] When the sampling condition holds, perform the following processing:

[0261] sample η trajectories from the trajectory set of U put them into

[0262] When the sampling condition does not hold, perform the following processing:

[0263] sample η trajectories from the trajectory set of U put them into

[0264] end

[0265] Regarding training the MetaMove model, the optimization objective of model training is shown in equation (13):

[0266]

[0267] where, denotes an optimizer that learns how to update the learner model 's parameters on the support set, which is also called the meta-learner. The parameters of the optimizer , and the parameters of the meta-learner , both coordinate with each other to achieve the optimal solution.

[0268] To train the meta move, the embodiments of the present application use a model-agnostic meta-learning optimization model to optimize the adaptability of the meta model , and the specific formula is shown in equation (14):

[0269]

[0270] where, denotes a specified hyperparameter representing the task learning rate, and L is the task-related loss calculated using the support set data . The model is trained to optimize the performance in all sampled tasks, and when the model is promoted to a new task, the representation of the model parameters becomes .

[0271] The parameter optimization process of the model is shown in equation (15):

[0272]

[0273] where, ​Let represent the meta-learning rate. The model learns based on gradient descent, as shown in formula (16):

[0274]

[0275]

[0276] The model training process is shown in the following pseudocode:

[0277] Input: Task learning rate α, meta-learning rate All tasks T and their associated datasets

[0278] Output: Model parameters .

[0279] Random initialization parameters

[0280] Perform the following processing:

[0281] Randomly sample a batch of tasks from all tasks T. , i = [l, B]

[0282] For each Perform the following processing:

[0283] When each sample exist When in transit, perform the following processing:

[0284] Calculation function Output

[0285] Calculate the loss function

[0286] Finish

[0287] use According to formula (14) in the support set To iteratively calculate the adjusted parameters

[0288] When each sample exist When in transit, perform the following processing:

[0289] Calculation function Output

[0290] Calculate the loss function

[0291] Finish

[0292] use According to formula (16) in the query set Calculate the partial derivatives

[0293] Update parameters using formula (15)

[0294] Finish

[0295] Finish

[0296] The training process described above follows purely supervised classification based on metadata, supporting labeled data in the support set. To further improve model performance, this embodiment utilizes unlabeled data to achieve performance enhancement. (The support set is then relaxed.) The requirement is to include a small number of unlabeled samples, specifically as follows: Figure 15 As shown, learning The trajectories with labels and the unlabeled trajectories in S' are used to infer the query set. The results above.

[0297] Considering that object mobility preferences are limited by geographical distance—meaning objects prefer to visit nearby Points of Interest (POIs)—and that most object activity is concentrated in a few areas, therefore... Figure 16 As shown, unlabeled data can be distinguished by calculating the Euclidean distance between unlabeled samples and the cluster centers of each known class, and the unlabeled data can be labeled with the most set of labels.

[0298] To verify the effects of the embodiments of this application, the following were performed: Figure 17 The effect comparison, by Figure 17 As can be seen, in terms of the click-through rate of the recommended information, the solution of this application embodiment improves the click-through rate by an average of 182.7% compared with other technical solutions. In terms of the conversion rate of the recommended information, the solution of this application embodiment improves the conversion rate by an average of 178.41% compared with other technical solutions.

[0299] The exemplary application and implementation of the electronic device provided in the embodiments of this application have thus far illustrated the artificial intelligence-based network request processing method provided in the embodiments of this application. The embodiments of this application also provide an artificial intelligence-based network request processing device. In practical applications, the functional modules in the artificial intelligence-based network request processing device can be collaboratively implemented by the hardware resources of the electronic device (such as a terminal, server, or server cluster), such as computing resources like processors, communication resources (such as those used to support various communication methods like optical fiber and cellular), and memory. Figure 2An artificial intelligence-based network request processing apparatus 555 stored in the memory 550 is shown, which can be software in the form of programs and plug-ins, etc., such as software modules designed in software programming languages such as C / C++, Java, etc., application software or special software modules in large software systems designed in software programming languages such as C / C++, Java, etc., application program interfaces, plug-ins, cloud services, etc. Different implementation manners are exemplified below.

[0300] The artificial intelligence-based network request processing apparatus 555 includes a series of modules, including an acquisition module 5551, a feature extraction module 5552, a classification module 5553, a prediction module 5554, and an application module 5555. The cooperation of each module in the artificial intelligence-based network request processing apparatus 555 provided by the embodiments of the present application to implement the network request processing scheme is described below.

[0301] The acquisition module 5551 is configured to acquire a feature data set of a target object and an action trajectory, wherein the action trajectory includes at least one historical interest point of a route; the feature extraction module 5552 is configured to perform feature extraction processing on the feature data set of the target object and the action trajectory to obtain an object feature of the target object for the action trajectory; the classification module 5553 is configured to perform trajectory classification processing on the action trajectory of the target object based on the object feature of the target object for the action trajectory to obtain a trajectory category to which the action trajectory belongs; the prediction module 5554 is configured to perform meta-learning-based interest point prediction processing based on the trajectory category to which the action trajectory belongs to obtain a target interest point to be passed through by the action trajectory; and the application module 5555 is configured to apply a network service mode matched with the target interest point to the target object.

[0302] In some embodiments, the feature extraction module 5552 is further configured to perform fusion processing on the feature data set of the target object and the action trajectory to obtain fusion information of the target object; and perform feature extraction processing on the fusion information of the target object to obtain the object feature of the target object for the action trajectory.

[0303] In some embodiments, the feature extraction processing is implemented by a feature extraction model, the feature extraction model comprising a plurality of cascaded feature extraction layers; the feature extraction module 5552 is further configured to perform feature extraction processing on the fusion information of the target object by a first feature extraction layer of the plurality of cascaded feature extraction layers; output the feature extraction result of the first feature extraction layer to a subsequent cascaded feature extraction layer, and continue to perform feature extraction processing and feature extraction result output by the subsequent cascaded feature extraction layer until output to a last feature extraction layer; and output the feature extraction result output by the last feature extraction layer as the object feature of the target object for the action trajectory.

[0304] In some embodiments, the feature extraction module 5552 is further configured to perform the following processing by a jth feature extraction layer of the plurality of cascaded feature extraction layers: perform random convolution processing on the feature extraction result of a (j-1)th feature extraction layer to obtain a jth deep layer feature; perform pooling processing on the jth deep layer feature to obtain the feature extraction result of the jth feature extraction layer; and output the feature extraction result of the jth feature extraction layer by the jth feature extraction layer; wherein j is a natural number and takes a value in a range of 1 < j ≤ N, N is the number of the feature extraction layers, and N is a positive integer greater than 1.

[0305] In some embodiments, the classification module 5553 is further configured to perform time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a time sequence feature of the action trajectory; and perform mapping processing on the time sequence feature of the action trajectory to obtain a trajectory category to which the action trajectory belongs.

[0306] In some embodiments, the classification module 5553 is further configured to perform forward time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a forward time sequence feature of the action trajectory; perform reverse time sequence encoding processing on the object feature of the target object for the action trajectory to obtain a reverse time sequence feature of the action trajectory; and perform splicing processing on the forward time sequence feature and the reverse time sequence feature to obtain the time sequence feature of the action trajectory.

[0307] In some embodiments, the classification module 5553 is further configured to perform, based on the i-1th forward time sequence feature, a forward screening processing on the object feature based on a forgetting gate to obtain a forward output feature of the forgetting gate; perform, based on the i-1th forward time sequence feature, a forward updating processing on the object feature based on an input gate to obtain a forward output feature of the input gate; perform, based on the i-1th forward time sequence feature, a forward mapping processing on the object feature based on an output gate to obtain a forward output feature of the output gate; and perform a nonlinear mapping processing on the forward output feature of the forgetting gate, the forward output feature of the input gate, and the forward output feature of the output gate to obtain an i-th forward time sequence feature; and take a set of M forward time sequence features as the forward time sequence features of the action trajectory; wherein i is a natural number that increases and takes a value in a range of 1 < i ≤ M, M is a feature quantity of the object feature, and M is a positive integer greater than 1.

[0308] In some embodiments, the classification module 5553 is further configured to perform, based on the e+1th reverse time sequence feature, a reverse screening processing on the object feature based on a forgetting gate to obtain a reverse output feature of the forgetting gate; perform, based on the e+1th reverse time sequence feature, a reverse updating processing on the object feature based on an input gate to obtain a reverse output feature of the input gate; perform, based on the e+1th reverse time sequence feature, a reverse mapping processing on the object feature based on an output gate to obtain a reverse output feature of the output gate; and perform a nonlinear mapping processing on the reverse output feature of the forgetting gate, the reverse output feature of the input gate, and the reverse output feature of the output gate to obtain an e-th reverse time sequence feature; and take a set of M reverse time sequence features as the reverse time sequence features of the action trajectory; wherein e is a natural number that decreases and takes a value in a range of 1 ≤ e ≤ M-1, M is a feature quantity of the object feature, and M is a positive integer greater than 1.

[0309] In some embodiments, the interest point prediction process is implemented by a meta-learning model, and a training process of the meta-learning model includes: constructing a query set and a support set based on a plurality of action trajectory samples, the support set including at least one support subset, each of the support subsets including at least one first action trajectory sample in the plurality of action trajectory samples, different support subsets including action trajectory samples of different trajectory categories, and the query set including at least one second action trajectory sample in the plurality of action trajectory samples; performing the following processes by an initialized meta-learning model: performing meta-learning-based interest point prediction processing on the first action trajectory samples included in the support subsets to obtain first predicted interest points to be passed through by the first action trajectory samples; performing meta-learning-based interest point prediction processing on the second action trajectory samples included in the query set to obtain second predicted interest points to be passed through by the second action trajectory samples; constructing a loss function of the meta-learning model based on the first predicted interest points, interest point labels corresponding to the first action trajectory samples, the second predicted interest points, and interest point labels corresponding to the second action trajectory samples; and updating parameters of the meta-learning model based on the loss function, and taking the updated parameters of the meta-learning model as parameters of the trained meta-learning model.

[0310] In some embodiments, the application module 5555 is further configured to query, based on the target interest point, a correspondence between different interest points and candidate network service modes, take a queried candidate network service mode as a network service mode matched with the target interest point, display recommendation information of the network service mode, and apply the network service mode to a terminal corresponding to the target object in response to an application operation for the network service mode based on the recommendation information.

[0311] In some embodiments, the application module 5555 is further configured to query, based on the target interest point, a correspondence between different interest points and candidate network service modes, take a queried candidate network service mode as a network service mode matched with the target interest point, and apply the network service mode matched with the target interest point to a terminal corresponding to the target object.

[0312] In some embodiments, the application module 5555 is further configured to acquire a distance from a terminal corresponding to the target object to the target interest point, and apply the network service mode matched with the target interest point to the terminal corresponding to the target object when the distance is less than a distance threshold.

[0313] In some embodiments, when a plurality of network service modes matched with the target point of interest exist, the application module 5555 is further configured to acquire a target time period when the target point of interest is reached; determine a network service mode matched with the target time period from the plurality of network service modes matched with the target point of interest; and cause the terminal application corresponding to the target object to use the network service mode matched with the target time period.

[0314] The embodiment of the present application provides a computer program product or a computer program, which comprises computer instructions stored in a computer readable storage medium. A processor of an electronic device reads the computer instructions from the computer readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the network request processing method based on artificial intelligence provided in the embodiment of the present application.

[0315] The embodiment of the present application provides a computer readable storage medium storing executable instructions, wherein the executable instructions are stored in the computer readable storage medium. When the executable instructions are executed by a processor, the processor will execute the network request processing method based on artificial intelligence provided in the embodiment of the present application, for example, as shown in the network request processing method based on artificial intelligence. Figures 3A-3C The network request processing method based on artificial intelligence is shown.

[0316] In some embodiments, the computer readable storage medium can be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disc, or CD-ROM; or various devices including one or any combination of the above memories.

[0317] In some embodiments, the executable instructions can be in the form of programs, software, software modules, scripts or codes, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and can be deployed in any form, including being deployed as independent programs or being deployed as modules, components, subroutines or other units suitable for use in a computing environment.

[0318] As an example, the executable instructions can but not necessarily correspond to files in a file system, can be stored in part of a file storing other programs or data, for example, stored in one or more scripts in a Hyper Text Markup Language (HTML) document, stored in a single file dedicated to the program in question, or stored in multiple cooperative files (for example, files storing one or more modules, subroutines or code portions).

[0319] As an example, the executable instructions can be deployed to execute on one computing device, or on multiple computing devices located at one site, or distributed across multiple sites and interconnected by a communication network.

[0320] The above merely provides illustration of the embodiment of the present application, but is not intended to limit the protection scope of the present application. Any modification, equivalent replacement, and improvement made within the spirit and range of the present application shall fall within the protection scope of the present application.

Claims

1. An artificial intelligence-based network request processing method, characterized by, The method comprises: acquiring a feature data set of a target object and a travel trajectory, wherein the travel trajectory comprises at least one historical interest point of a route; performing feature extraction processing on the feature data set of the target object and the travel trajectory to obtain an object feature of the target object for the travel trajectory; performing trajectory classification processing on the travel trajectory of the target object based on the object feature of the target object for the travel trajectory to obtain a trajectory category to which the travel trajectory belongs; performing meta-learning-based interest point prediction processing based on the trajectory category to which the travel trajectory belongs to obtain a target interest point to be passed through by the travel trajectory; querying a correspondence between different interest points and candidate network service modes based on the target interest point, and taking the queried candidate network service mode as a network service mode matched with the target interest point; when the network service mode matched with the target interest point is multiple, acquiring a target time period when the target object arrives at the target interest point; from the multiple network service modes matched with the target interest point, determining a network service mode matched with the target time period; applying the network service mode matched with the target time period to a terminal application corresponding to the target object.

2. The method of claim 1, wherein, The trajectory classification processing on the travel trajectory of the target object based on the object feature of the target object for the travel trajectory to obtain a trajectory category to which the travel trajectory belongs comprises: performing time sequence coding processing on the object feature of the target object for the travel trajectory to obtain a time sequence feature of the travel trajectory; performing mapping processing on the time sequence feature of the travel trajectory to obtain the trajectory category to which the travel trajectory belongs.

3. The method of claim 2, wherein, The time sequence coding processing on the object feature of the target object for the travel trajectory to obtain a time sequence feature of the travel trajectory comprises: performing forward time sequence coding processing on the object feature of the target object for the travel trajectory to obtain a forward time sequence feature of the travel trajectory; performing reverse time sequence coding processing on the object feature of the target object for the travel trajectory to obtain a reverse time sequence feature of the travel trajectory; performing splicing processing on the forward time sequence feature and the reverse time sequence feature to obtain the time sequence feature of the travel trajectory.

4. The method of claim 3, wherein, The forward time sequence coding processing on the object feature of the target object for the travel trajectory to obtain a forward time sequence feature of the travel trajectory comprises: performing forward filtering processing on the object feature based on a forgetting gate based on an i-1th forward time sequence feature to obtain a forward output feature of the forgetting gate; performing forward updating processing on the object feature based on an input gate based on the i-1th forward time sequence feature to obtain a forward output feature of the input gate; performing forward mapping processing on the object feature based on an output gate based on the i-1th forward time sequence feature to obtain a forward output feature of the output gate; performing nonlinear mapping processing on the forward output feature of the forgetting gate, the forward output feature of the input gate, and the forward output feature of the output gate to obtain an i th forward time sequence feature; Taking the set of M positive temporal features as the positive temporal features of the action trajectory; where i is an increasing natural number and the value range is 1 < i ≤ M, M is the number of features of the object feature, and M is a positive integer greater than 1.

5. The method of claim 3, wherein, The reverse temporal encoding process for the object feature of the target object with respect to the action trajectory to obtain the reverse temporal feature of the action trajectory includes: Performing a reverse screening process based on the forgetting gate on the object feature based on the (e + 1)-th reverse temporal feature to obtain the reverse output feature of the forgetting gate; Performing a reverse update process based on the input gate on the object feature based on the (e + 1)-th reverse temporal feature to obtain the reverse output feature of the input gate; Performing a reverse mapping process based on the output gate on the object feature based on the (e + 1)-th reverse temporal feature to obtain the reverse output feature of the output gate; Performing a non-linear mapping process on the reverse output feature of the forgetting gate, the reverse output feature of the input gate, and the reverse output feature of the output gate to obtain the e-th reverse temporal feature; Taking the set of M reverse temporal features as the reverse temporal features of the action trajectory; where e is a decreasing natural number and the value range is 1 ≤ e ≤ M - 1, M is the number of features of the object feature, and M is a positive integer greater than 1.

6. The method of claim 1, wherein, The feature extraction process for the feature data set and action trajectory of the target object to obtain the object feature of the target object with respect to the action trajectory includes: Performing a fusion process on the feature data set and action trajectory of the target object to obtain the fusion information of the target object; Performing a feature extraction process on the fusion information of the target object to obtain the object feature of the target object with respect to the action trajectory.

7. The method according to claim 6, wherein the feature extraction process is implemented by a feature extraction model, and the feature extraction model includes multiple cascaded feature extraction layers; The feature extraction process for the fusion information of the target object to obtain the object feature of the target object with respect to the action trajectory includes: Performing a feature extraction process on the fusion information of the target object through the first feature extraction layer in the multiple cascaded feature extraction layers; Outputting the feature extraction result of the first feature extraction layer to the subsequent cascaded feature extraction layers, and continuously performing a feature extraction process and outputting the feature extraction result through the subsequent cascaded feature extraction layers until it is output to the last feature extraction layer; Taking the feature extraction result output by the last feature extraction layer as the object feature of the target object with respect to the action trajectory.

8. The method of claim 7, wherein, The continuous feature extraction process and output of the feature extraction result through the subsequent cascaded feature extraction layers include: Performing the following process through the j-th feature extraction layer in the multiple cascaded feature extraction layers: Performing a random convolution process on the feature extraction result of the (j - 1)-th feature extraction layer to obtain the j-th deep feature; Performing a pooling process on the j-th deep feature to obtain the feature extraction result of the j-th feature extraction layer; output a feature extraction result of the jth feature extraction layer by the jth feature extraction layer; wherein j is an increasing natural number and takes a value range of 1 < j ≤ N, N is the number of the feature extraction layers, and N is a positive integer greater than 1.

9. The method of claim 1, wherein, The interest point prediction processing is implemented by a meta-learning model, and a training process of the meta-learning model includes: constructing a query set and a support set based on a plurality of action trajectory samples, the support set including at least one support subset, each support subset including at least one first action trajectory sample in the plurality of action trajectory samples, different support subsets including action trajectory samples of different trajectory categories, and the query set including at least one second action trajectory sample in the plurality of action trajectory samples; performing the following processing by the initialized meta-learning model: performing meta-learning-based interest point prediction processing on the first action trajectory samples included in the support subset to obtain first predicted interest points to be passed through by the first action trajectory samples; performing meta-learning-based interest point prediction processing on the second action trajectory samples included in the query set to obtain second predicted interest points to be passed through by the second action trajectory samples; constructing a loss function of the meta-learning model based on the first predicted interest points, interest point labels corresponding to the first action trajectory samples, the second predicted interest points, and interest point labels corresponding to the second action trajectory samples; updating parameters of the meta-learning model based on the loss function, and taking the updated parameters of the meta-learning model as parameters of the trained meta-learning model.

10. The method of claim 1, wherein, The method further includes: querying a correspondence between different interest points and candidate network service modes based on the target interest point, and taking a queried candidate network service mode as a network service mode matched with the target interest point; displaying recommendation information of the network service mode; in response to an application operation for the network service mode based on the recommendation information, applying the network service mode to a terminal corresponding to the target object.

11. The method of claim 1, wherein, The method further includes: obtaining a distance from a terminal corresponding to the target object to the target interest point; when the distance is less than a distance threshold, applying a network service mode matched with the target interest point to the terminal corresponding to the target object.

12. An artificial intelligence-based network request processing apparatus, characterized by comprising: The apparatus includes: an acquisition module configured to acquire a feature data set of a target object and an action trajectory, wherein the action trajectory includes at least one passed historical interest point; a feature extraction module configured to perform feature extraction processing on the feature data set of the target object and the action trajectory to obtain an object feature of the target object for the action trajectory; a classification module configured to perform trajectory classification processing on the action trajectory of the target object based on the object feature of the target object for the action trajectory to obtain a trajectory category to which the action trajectory belongs; a prediction module configured to perform meta-learning-based interest point prediction processing based on the trajectory category to which the action trajectory belongs to obtain a target interest point to be passed through by the action trajectory; and a display module configured to display the target interest point. The application module is configured to query a correspondence between different interest points and candidate network service modes based on the target interest point, and take the queried candidate network service mode as a network service mode matched with the target interest point. When the network service mode matched with the target interest point is multiple, a target time period when a target object reaches the target interest point is acquired. From the multiple network service modes matched with the target interest point, a network service mode matched with the target time period is determined. The target object corresponds to a terminal application, and the network service mode matched with the target time period is applied.

13. The apparatus of claim 12, wherein, The classification module is further configured to: perform time sequence coding processing on the object feature of the target object for the action trajectory to obtain a time sequence feature of the action trajectory, and perform mapping processing on the time sequence feature of the action trajectory to obtain a trajectory category to which the action trajectory belongs.

14. The apparatus of claim 13, wherein, The classification module is further configured to: perform forward time sequence coding processing on the object feature of the target object for the action trajectory to obtain a forward time sequence feature of the action trajectory; perform reverse time sequence coding processing on the object feature of the target object for the action trajectory to obtain a reverse time sequence feature of the action trajectory; perform splicing processing on the forward time sequence feature and the reverse time sequence feature to obtain a time sequence feature of the action trajectory.

15. The apparatus of claim 14, wherein, The classification module is further configured to: perform forward filtering processing on the object feature based on a forgetting gate based on an i-1th forward time sequence feature to obtain a forward output feature of the forgetting gate, perform forward updating processing on the object feature based on an input gate based on the i-1th forward time sequence feature to obtain a forward output feature of the input gate, perform forward mapping processing on the object feature based on an output gate based on the i-1th forward time sequence feature to obtain a forward output feature of the output gate, and perform nonlinear mapping processing on the forward output feature of the forgetting gate, the forward output feature of the input gate, and the forward output feature of the output gate to obtain an i-th forward time sequence feature; a set of M forward time sequence features is taken as the forward time sequence feature of the action trajectory; wherein i is a natural number that increases and takes a value range of 1 16. The apparatus of claim 14, wherein, The classification module is further configured to: The object feature is subjected to a forgetting gate-based reverse screening process based on the e+1th reverse time sequence feature, to obtain a reverse output feature of the forgetting gate; the object feature is subjected to an input gate-based reverse update process based on the e+1th reverse time sequence feature, to obtain a reverse output feature of the input gate; the object feature is subjected to an output gate-based reverse mapping process based on the e+1th reverse time sequence feature, to obtain a reverse output feature of the output gate; the reverse output feature of the forgetting gate, the reverse output feature of the input gate, and the reverse output feature of the output gate are subjected to a nonlinear mapping process, to obtain the e th reverse time sequence feature; a set of M reverse time sequence features is taken as the reverse time sequence feature of the action trajectory; wherein e is a natural number in descending order and takes a value in a range of 1≤e≤M-1, M is a feature quantity of the object feature, and M is a positive integer greater than 1.

17. The apparatus of claim 12, wherein, The classification module is further configured to: fuse the feature data set of the target object and the action trajectory to obtain fusion information of the target object; and extract an object feature of the target object for the action trajectory from the fusion information of the target object.

18. An electronic device, comprising: The electronic device comprises: a memory configured to store executable instructions; a processor configured to execute the executable instructions stored in the memory, and implement the network request processing method based on artificial intelligence according to any one of claims 1 to 11.

19. A computer-readable storage medium, characterized in that, The executable instructions are stored in the memory and are configured to be executed by the processor to implement the network request processing method based on artificial intelligence according to any one of claims 1 to 11.

20. A computer program product comprising computer programs or instructions, characterized in that, The computer program or instructions are executed by the processor to implement the network request processing method based on artificial intelligence according to any one of claims 1 to 11.

Images