Network optimization for a cell

Abstract

Embodiments of the present disclosure provide a solution for network optimization. A method for network optimization comprises: generating a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell; classifying, based on feature information of nodes and edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types; determining for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and determining for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell. Accordingly, automated and accurate network optimization recommendation may be achieved.

Description

NETWORK OPTIMIZATION FOR A CELLFIELD

[0001] Embodiments of the present disclosure generally relate to the field of telecommunication and in particular, to network optimization for a cell.BACKGROUND

[0002] Wireless network optimization may improve the performance of cells. In the practice of wireless network optimization, some poor performance on cells may be observed, and network engineers need to analyze which type of issue the cell is suffering and take some operations accordingly to improve the performance of the cell. In most common use cases, the analysis is done manually and depends on the knowledge of the engineers. As the wireless network is becoming bigger and more complicated, the way of manually diagnosing by engineers is less efficient. There is a need for automated solutions to solve this problem in a more efficient and accurate way.

[0003] For defining the type of issue, there are some related solutions for network fault detection. In one solution, it applies clustering techniques to identify clusters with similar performance issues based on key performance indicator (KPI) data. In another solution, Autoencoder is used for detecting the fault nodes among a large amount of network nodes. Their works mostly focus on defining the issue type of the single network node. For single node solution above, it is obvious that this kind of solutions ignores the connection between cells and its neighbors. But from the practice of issue cell analysis, the impact of its neighbors is very important, which helps to provide a more accurate operation of network optimization. Their works mostly focus on defining the issue type of the cell and do not move forward to the interaction between cells and neighbors.

[0004] Usually, the cause of the poor performance doesn’t exist in the cell itself (also referred to as serving cell), which means the neighbor cells also contribute to the performance and need to be analyzed, and the network tuning approaches can be implemented on both the issue cell and its neighbors. In some cases, the engineers firstly locate the issue type of the cell, and then collect the data of its neighbors. Based on both data from the cell and its neighbors, engineers diagnose where the poor performance come from, and it may be caused by the inappropriate settings of the cell itself or its neighbor cells. Then these cells are marked as target cells, where a target cell represents a cell to be adjusted for the settings in the wireless network. However, such a manual, timeconsuming process is not desirable, leading to low efficiency.SUMMARY

[0005] As the wireless network are typically represented as graphs with complex relationships and interdependency between network nodes, there are some solutions aiming at defining the issue type and finding the target nodes in a graph. Graph Neural Networks (GNN) techniques provide a new way to solve this kind of problem based on the great power to deal with the dependency between measurements. For example, in some solutions, the device network is set up and a GNN model is built to diagnose the fault of all devices and locate the fault-root-devices. In some other solutions, the root cause identification problem is formulated as a node level classification problem. Based on the node performance, which is represented by a vector of KPIs, GNN is adopted to predict root cause cells over wireless networks. All these solutions are mostly applying node classification on a graph to give a fault label to each node.

[0006] As the wireless network can be represented as graphs, for above solutions of node classification on graph, all the nodes are defined as different types of fault nodes, but it is not clear which node is an issue node and which node contributes to the issue and which works as a target node. Because the cause of the poor performance of the cells can be various, the issue type is usually not restricted to only one fault label, the existing solutions cannot decompose the contributions of each node to the issue. Also, these methods do not differentiate distinct type of nodes, but in practice the type of each node (corresponding to each cell) is always different, and the features considered for distinct type of nodes can be different as well, which means the graph structure is heterogeneous instead of homogeneous. Additionally defining labels of each node on a big graph is also a tough work in network operation. Therefore, although the GNN methods provides a new insight, it still needs to move forward considering the demands for solving these problems.

[0007] Considering the demands for GNN methods and an explainable solution to identify the impact of each node to the classification result, the method used to explain the model is often referred to as explainable Artificial Intelligence (XAI) in neural networks. In the GNN domain, there is initial trial of Graph Convolutional Network (GCN) with Gradient-weighted Class Activation Maps (Grad-CAM) which is used in the classification and interpretation of brain connectivity graph. However, this method ignores the impact of message passing in GCN which could affect the performance of Grad-CAM.

[0008] Thus, an automated and efficient solution of network optimization is needed.

[0009] Embodiments of the present disclosure propose a method of implementing end-to-end subgraph network optimization action type classification and target nodes identification. The output of the method provides optimization recommendations which include network optimization action types (adjusting operations and corresponding parameters) and the target nodes to implement these actions.

[0010] In a first aspect of the present disclosure, a computer-implemented method for network optimization is proposed. The method comprises: generating a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes; classifying, based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell; determining for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and determining for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell. Through the subgraph-based classification and target cell identification, automated and accurate network optimization recommendation may be performed for the target cell(s) that has impact on the poor performance of the serving cell.

[0011] In some embodiments of the first aspect, the nodes in the subgraph comprise: a first node with a first node type representing the serving cell, the first node type indicating that the serving cell is an issue cell, and at least one second node with a second node type representing the at least one neighbor cell, the second node type indicating a neighbor cell of the issue cell; and wherein a type of feature information of the first node is different from a type of feature information of the at least one second node.

[0012] In some embodiments of the first aspect, generating a subgraph for the serving cell comprises: extracting, based on the serving cell, the subgraph from a graph corresponding to the communication network, the graph comprising a set of nodes and edges connecting the set of nodes, the set of nodes representing respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

[0013] In some embodiments of the first aspect, classifying the subgraph into at least one of a plurality of network optimization action types comprises: performing a subgraph classification on the subgraph using a GNN, the GNN comprising a plurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph; obtaining an output of the GNN, the output of the GNN at least comprising a plurality of labels corresponding to the plurality of network optimization action types, each of the plurality of labels indicating a predicted probability of the subgraph being classified into a corresponding network optimization action type;and determining the at least one network optimization action type for the subgraph based on the output of the GNN.

[0014] In some embodiments of the first aspect, the GNN at least comprises one or more of: a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing the at least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, the intermediate feature representation being determined based on the feature information of the first node and the at least one second node.

[0015] In some embodiments of the first aspect, determining the at least one target cell for each of the at least one network optimization action type comprises: for a given network optimization action type among the at least one network optimization action type, extracting an intermediate feature representation of the feature information of the nodes and the edges in the subgraph from a GNN, wherein the feature information is provided as input to the GNN for processing, and the classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation; determining respective contribution scores of the nodes in the subgraph to the given network optimization action type based on the intermediate feature representation, a contribution score of a node indicating a contribution of a cell represented by the node to the given network optimization action type; and determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes.

[0016] In some embodiments of the first aspect, the GNN comprising a plurality of processing layers, and the intermediate feature representation is extracted from an output of a processing layer of the GNN.

[0017] In some embodiments of the first aspect, determining respective contribution scores of the nodes in the subgraph to the given network optimization action type comprises: determining, based on the intermediate feature representation, respective weights of the nodes in the subgraph with respect to the given network optimization action type; applying a transformation operation on a weighted intermediate feature representation, the weighted intermediate feature representation being obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation; and determining respective contribution scores of the nodes to the given network optimization action type based on the transformed intermediate feature representation.

[0018] In some embodiments of the first aspect, the transformed intermediate feature representation comprises non-normalized contribution scores of the nodes to the given network optimization action type, and wherein determining respective contribution scores of the nodes based on the transformed intermediate feature representation comprises: adjusting the nonnormalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and applying a normalization operation on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.

[0019] In some embodiments of the first aspect, adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell comprises: in accordance with a determination that the given network optimization action type is associated with the at least one neighbor cell, replacing a non-normalized contribution score of a first node representing the serving cell with a score of zero; and in accordance with a determination that the given network optimization action type is associated with the serving cell, replacing at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell with a score of zero.

[0020] In some embodiments of the first aspect, determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes comprises: comparing the respective contribution scores of the nodes with a contribution threshold for the given network optimization action type; selecting at least one of the nodes with a contribution score exceeding the contribution threshold; and determining at least one cell represented by the at least one selected node in the subgraph as the at least one target cell for the given network optimization action type.

[0021] In some embodiments of the first aspect, the method further comprises: determining the contribution threshold for the given network optimization action type by: obtaining a set of sample subgraphs classified into the given network optimization action type, and a set of groundtruth target cells in the set of sample subgraphs for the given network optimization action type, wherein the set of sample subgraphs are sampled from a training dataset for the GNN; determining, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given network optimization action type with a plurality of candidate contribution thresholds; determining respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells; and selecting, from the plurality of candidate contribution thresholds, the contributionthreshold for the given network optimization action type based on the respective matching ratios of the plurality of sets of candidate target cells.

[0022] In some embodiments of the first aspect, the feature information of a node in the subgraph indicates at least one of the following: key performance indicator information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node; and / or wherein the feature information of an edge between two nodes in the subgraph indicates a strength of the relationship between two cells represented by the two nodes.

[0023] In some embodiments of the first aspect, the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

[0024] In a second aspect, an apparatus for network optimization is proposed. The apparatus comprises means for generating a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes; means for classifying, based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell; means for determining for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and means for determining for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell.

[0025] In a third aspect, a computer readable medium is proposed. The computer readable medium has instructions stored thereon, the instructions when executed by at least one processor causing the at least one processor to perform the method according to any of the embodiments of the first aspect.

[0026] In a fourth aspect, a computer program product is proposed. The computer program product comprises instructions which, when executed on at least one processor, cause the at least one processor to carry out the method according to any of the embodiments of the first aspect.

[0027] According to the various embodiments of the present disclosure, a subgraph is generated to define the network optimization action type(s) and the target cell(s), and thus the network optimization recommendations can be automated based on the recognized action type(s)and target cell(s). In addition, the GNN model can support heterogenous input and multi-label outputs for the classification of the network optimization action types. Contribution scores can be produced to indicate the contribution of a node in the subgraph to each network optimization type, and thus can be used to identify the target nodes to be optimized. Further, the contribution accuracy can be further improved by restricting the target node types (serving node or neighbor nodes) to calculate the contribution scores per network optimization action type. In some case, the GNN model can be utilized to achieve end-to-end subgraph classification and target node identification, which improve the efficiency and cause reduced consumption.

[0028] It is to be understood that the Summary section is not intended to identify key or essential features of embodiments of the present disclosure, nor is it intended to be used to limit the scope of the present disclosure. Other features of the present disclosure will become easily comprehensible through the following description.BRIEF DESCRIPTION OF THE DRAWINGS

[0029] Through the more detailed description of some example embodiments of the present disclosure in the accompanying drawings, the above and other objects, features and advantages of the present disclosure will become more apparent, where:

[0030] FIG. 1 illustrates an example of a communication system in which some embodiments of the present disclosure can be implemented;

[0031] FIG. 2 illustrates a flowchart of a process for network optimization in accordance with some embodiments of the present disclosure;

[0032] FIG. 3 illustrates an example of different stages at the process of network optimization in accordance with some further embodiments of the present disclosure;

[0033] FIG. 4 illustrates a schematic diagram of training and application stages of GNN-based network optimization in accordance with some embodiments of the present disclosure;

[0034] FIG. 5 illustrates an example of multi-label prediction by the GNN in accordance with some further embodiments of the present disclosure;

[0035] FIG. 6 illustrates a block diagram of an example model structure of the GNN in accordance with some embodiments of the present disclosure;

[0036] FIG. 7 illustrates a schematic diagram of an example of contribution score determination in accordance with some embodiments of the present disclosure;

[0037] FIG. 8 illustrates flowcharts of different flows for contribution score determination in accordance with some embodiments of the present disclosure;

[0038] FIG. 9 illustrates a block diagram of an example cloud environment in accordance with some embodiments of the present disclosure;

[0039] FIG.10 illustrates a flowchart of a computer-implemented method for network optimization in accordance with some embodiments of the present disclosure;

[0040] FIG. 11 is a block diagram showing an apparatus suitable for use in practicing some embodiments of the present disclosure;

[0041] FIG. 12 is a block diagram of a host which may be an embodiment of the host of FIG. 1 in accordance with some embodiments of the present disclosure;

[0042] FIG. 13 is a block diagram illustrating a virtualization environment in which functions implemented by some embodiments of the present disclosure may be virtualized; and

[0043] FIG. 14 shows a communication diagram of a host communicating via a network node with a UE over a partially wireless connection in accordance with some embodiments of the present disclosure.

[0044] Throughout the drawings, the same or similar reference numerals represent the same or similar element.DETAILED DESCRIPTION

[0045] Some of the embodiments contemplated herein will now be described more fully with reference to the accompanying drawings. The embodiments set forth below represent information to enable those skilled in the art to practice the embodiments and illustrate the best mode of practicing the embodiments. Upon reading the following description in light of the accompanying figures, those skilled in the art will understand the concepts of the disclosure and will recognize applications of these concepts not particularly addressed herein. It should be understood that these concepts and applications fall within the scope of the disclosure.

[0046] Generally, all terms used herein are to be interpreted according to their ordinary meaning in the relevant technical field, unless a different meaning is clearly given and / or is implied from the context in which it is used. All references to a / an / the element, apparatus, component, means, step, etc. are to be interpreted openly as referring to at least one instance of the element, apparatus, component, means, step, etc., unless explicitly stated otherwise. The steps of any methods disclosed herein do not have to be performed in the exact order disclosed, unless a step is explicitly described as following or preceding another step and / or where it is implicit that a step must follow or precede another step. Any feature of any of the embodiments disclosed herein may be applied to any other embodiment, wherever appropriate. Likewise, any advantage of any of the embodiments may apply to any other embodiments, and vice versa. Other objectives, features, and advantages of the enclosed embodiments will be apparent from the following description.

[0047] As used herein, the term “includes” and its variants are to be read as open terms that mean “includes, but is not limited to.” The term “based on” is to be read as “at least in part based on.” The term “one embodiment” and “an embodiment” are to be read as “at least oneembodiment.” The term “another embodiment” is to be read as “at least one other embodiment.” The terms “first,” “second,” and the like may refer to different or same objects. Other definitions, explicit and implicit, may be included below.

[0048] As used herein, the term “terminal device” refers to any device having wireless or wired communication capabilities. Examples of the terminal device include, but not limited to, user equipment (UE), personal computers, desktops, mobile phones, cellular phones, smart phones, personal digital assistants (PDAs), portable computers, tablets, wearable devices, internet of things (loT) devices, Ultra-reliable and Low Latency Communications (URLLC) devices, Internet of Everything (loE) devices, machine type communication (MTC) devices, devices on vehicle for V2X communication where X means pedestrian, vehicle, or infrastructure / network, devices for Integrated Access and Backhaul (IAB), Space borne vehicles or Air borne vehicles in Nonterrestrial networks (NTN) including Satellites and High Altitude Platforms (HAPs) encompassing Unmanned Aircraft Systems (UAS), extended Reality (XR) devices including different types of realities such as Augmented Reality (AR), Mixed Reality (MR) and Virtual Reality (VR), the unmanned aerial vehicle (UAV) commonly known as a drone which is an aircraft without any human pilot, devices on high speed train (HST), or image capture devices such as digital cameras, sensors, gaming devices, music storage and playback appliances, or Internet appliances enabling wireless or wired Internet access and browsing and the like. The “terminal device” can further have “multicast / broadcasf ’ feature, to support public safety and mission critical, V2X applications, transparent IPv4 / IPv6 multicast delivery, IPTV, smart TV, radio services, software delivery over wireless, group communications and loT applications. It may also incorporate one or multiple Subscriber Identity Module (SIM) as known as Multi-SIM. The term “terminal device” can be used interchangeably with a UE, a mobile station, a subscriber station, a mobile terminal, a user terminal or a wireless device.

[0049] As used herein, the term “network node” may also be referred to as a network function (NF), a network entity, or a network device, and refers to a physical, virtual or hybrid function or entity which is deployed at a network side and provides one or more services to clients / consumers. For example, an NF may be arranged at a device in a radio access network (RAN) or a core network (CN) of a communication system. The network node may be implemented in hardware, software, firmware, or some combination thereof. Examples of a network node in a RAN include, but not limited to, a Node B (NodeB or NB), an evolved NodeB (eNodeB or eNB), a next generation NodeB (gNB), a transmission reception point (TRP), a remote radio unit (RRU), a radio head (RH), a remote radio head (RRH), an IAB node, a low power node such as a femto node, a pico node, a reconfigurable intelligent surface (RIS), and the like. Examples of a network node in a CN include, but not limited to, a Mobility Management Entity (MME), a Packet Data Network Gateway (P-GW), a Service Capability Exposure Function (SCEF), a Home Subscriber Server (HSS), or the like. Some other examples of a core network node include a node implementing an Access and Mobility Management Function (AMF), a User Plane Function (UPF), a Session Management Function (SMF), an Authentication Server Function (AUSF), a Network Slice Selection Function (NSSF), a Network Exposure Function (NEF), a Network Function (NF) Repository Function (NRF), a Policy Control Function (PCF), a Unified Data Management (UDM), or the like.

[0050] Communications in a communication environment may conform to any suitable standards including, but not limited to, Global System for Mobile Communications (GSM), Long Term Evolution (LTE), LTE-Evolution, LTE-Advanced (LTE-A), New Radio (NR), Wideband Code Division Multiple Access (WCDMA), Code Division Multiple Access (CDMA), GSM EDGE Radio Access Network (GERAN), Machine Type Communication (MTC) and the like. The embodiments of the present disclosure may be performed according to any generation communication protocols either currently known or to be developed in the future. Examples of the communication protocols include, but not limited to, the first generation (1G), the second generation (2G), 2.5G, 2.75G, the third generation (3G), the fourth generation (4G), 4.5G, the fifth generation (5G) communication protocols, 5.5G, 5G-Advanced networks, or the sixth generation (6G) networks and beyond and the like, wireless local network communication protocols such as Institute for Electrical and Electronics Engineers (IEEE) 802.11 and the like, and / or any other protocols currently known or to be developed in the future. Moreover, the communication may utilize any proper wireless communication technology, comprising but not limited to: Code Division Multiple Access (CDMA), Frequency Division Multiple Access (FDMA), Time Division Multiple Access (TDMA), Frequency Division Duplex (FDD), Time Division Duplex (TDD), Multiple-Input Multiple-Output (MIMO), Orthogonal Frequency Division Multiple (OFDM), Discrete Fourier Transform spread OFDM (DFT-s-OFDM) and / or any other technologies currently known or to be developed in the future.

[0051] As used herein, the term “model” is referred to as an association between an input and an output learned from training data, and thus a corresponding output may be generated for a given input after the training. The generation of the model may be based on machine learning (ML) techniques. The machine learning techniques may also be referred to as artificial intelligence (Al) techniques. In general, a machine learning model can be built, which receives input information and makes predictions based on the input information. For example, a classification model may predict a class of the input information among a predetermined set of classes. As used herein, “model” may also be referred to as “machine learning model”, “learning model”, “machine learning network”, or “learning network,” which are used interchangeably herein.

[0052] Generally, machine learning may usually involve three stages, i.e., a training stage, a validation stage, and an application stage (also referred to as an inference stage). At the training stage, a given machine learning model may be trained (or optimized) iteratively using a great amount of training data until the model can obtain, from the training data, consistent inference similar to those that human intelligence can make. During the training, a set of parameter values of the model is iteratively updated until a training objective is reached. Through the training process, the machine learning model may be regarded as being capable of learning the association between the input and the output (also referred to an input-output mapping) from the training data. At the validation stage, a validation input is applied to the trained machine learning model to test whether the model can provide a correct output, so as to determine the performance of the model. Generally, the validation stage may be considered as a step in a training process, or sometimes may be omitted. At the inference stage, the trained machine learning model may be used to process a real-world model input based on the set of parameter values obtained from the training process and to determine the corresponding model output.

[0053] FIG. 1 shows an example of a communication system 100 in which some embodiments of the present disclosure can be implemented. In the example, the communication system 100 includes a telecommunication network 102 that includes an access network 104, such as a radio access network (RAN), and a core network 106, which includes one or more core network nodes 108. The access network 104 includes one or more access network nodes, such as network nodes 110a and 110b (one or more of which may be generally referred to as network nodes 110), or any other similar 3rdGeneration Partnership Project (3GPP) access node or non-3GPP access point. The network nodes 110 facilitate direct or indirect connection of user equipment (UE), such as by connecting terminal devices 112a, 112b, 112c, and 112d (one or more of which may be generally referred to as terminal devices 112) to the core network 106 over one or more wireless connections.

[0054] A network optimization device 120 may be included in the communication system 100 for network performance optimization. The network optimization device 120 may be a separate device, such as a core network node 108, a network node 110, etc. In some cases, the network optimization device 120 may be integrated into network devices or the network optimization method disclosed herein may be implemented by the network devices. For ease of description, the following will be described from the perspective of the network optimization device 120.

[0055] Example wireless communications over a wireless connection include transmitting and / or receiving wireless signals using electromagnetic waves, radio waves, infrared waves, and / or other types of signals suitable for conveying information without the use of wires, cables, or other material conductors. Moreover, in different embodiments, the communication system 100 may include any number of wired or wireless networks, network nodes, terminal devices, and / or anyother components or systems that may facilitate or participate in the communication of data and / or signals whether via wired or wireless connections. The communication system 100 may include and / or interface with any type of communication, telecommunication, data, cellular, radio network, and / or other similar type of system.

[0056] The terminal devices 112 may be any of a wide variety of communication devices, including wireless devices arranged, configured, and / or operable to communicate wirelessly with the network nodes 110 and other communication devices. Similarly, the network nodes 110 are arranged, capable, configured, and / or operable to communicate directly or indirectly with the terminal devices 112 and / or with other network nodes or equipment in the telecommunication network 102 to enable and / or provide network access, such as wireless network access, and / or to perform other functions, such as administration in the telecommunication network 102.

[0057] In the depicted example, the core network 106 connects the network nodes 110 to one or more hosts, such as host 116. These connections may be direct or indirect via one or more intermediary networks or devices. In other examples, network nodes may be directly coupled to hosts. The core network 106 includes one or more core network nodes (e.g., one or more core network nodes 108) that are structured with hardware and software components. Features of these components may be substantially similar to those described with respect to the terminal devices, network nodes, and / or hosts, such that the descriptions thereof are generally applicable to the corresponding components of the core network node 108. Example core network nodes include functions of one or more of a Mobile Switching Center (MSC), Mobility Management Entity (MME), Home Subscriber Server (HSS), Access and Mobility Management Function (AMF), Session Management Function (SMF), Authentication Server Function (AUSF), Subscription Identifier De-concealing function (SIDF), Unified Data Management (UDM), Security Edge Protection Proxy (SEPP), Network Exposure Function (NEF), and / or a User Plane Function (UPF).

[0058] The host 116 may be under the ownership or control of a service provider other than an operator or provider of the access network 104 and / or the telecommunication network 102 and may be operated by the service provider or on behalf of the service provider. The host 116 may host a variety of applications to provide one or more service. Examples of such applications include live and pre-recorded audio / video content, data collection services such as retrieving and compiling data on various ambient conditions detected by a plurality of terminal devices, analytics functionality, social media, functions for controlling or otherwise interacting with remote devices, functions for an alarm and surveillance center, or any other such function performed by a server.

[0059] As a whole, the communication system 100 of FIG. 1 enables connectivity between the terminal devices, network nodes, and hosts. In that sense, the communication system may be configured to operate according to predefined rules or procedures, such as specific standards thatinclude, but are not limited to: Global System for Mobile Communications (GSM); Universal Mobile Telecommunications System (UMTS); Long Term Evolution (LTE), and / or other suitable 2G, 3G, 4G, 5G standards, or any applicable future generation standard (e.g., 6G); wireless local area network (WLAN) standards, such as the Institute of Electrical and Electronics Engineers (IEEE) 802.11 standards (WiFi); and / or any other appropriate wireless communication standard, such as the Worldwide Interoperability for Microwave Access (WiMax), Bluetooth, Z-Wave, Near Field Communication (NFC) ZigBee, LiFi, and / or any low-power wide-area network (LPWAN) standards such as LoRa and Sigfox.

[0060] In some examples, the telecommunication network 102 is a cellular network that implements 3GPP standardized features. Accordingly, the telecommunications network 102 may support network slicing to provide different logical networks to different devices that are connected to the telecommunication network 102. For example, the telecommunications network 102 may provide Ultra Reliable Low Latency Communication (URLLC) services to some terminal devices, while providing Enhanced Mobile Broadband (eMBB) services to other terminal devices, and / or Massive Machine Type Communication (mMTC) / Massive loT services to yet further terminal devices.

[0061] In some examples, the terminal devices 112 are configured to transmit and / or receive information without direct human interaction. For instance, a terminal device may be designed to transmit information to the access network 104 on a predetermined schedule, when triggered by an internal or external event, or in response to requests from the access network 104. Additionally, a terminal device may be configured for operating in single- or multi-RAT or multi-standard mode. For example, a terminal device may operate with any one or combination of Wi-Fi, NR (New Radio) and LTE, i.e., being configured for multi-radio dual connectivity (MR-DC), such as E- UTRAN (Evolved-UMTS Terrestrial Radio Access Network) New Radio - Dual Connectivity (EN-DC).

[0062] In the example, the hub 114 communicates with the access network 104 to facilitate indirect communication between one or more terminal devices (e.g., terminal device 112c and / or 112d) and network nodes (e.g., network node 110b). In some examples, the hub 114 may be a controller, router, content source and analytics, or any of the other communication devices described herein regarding terminal devices. For example, the hub 114 may be a broadband router enabling access to the core network 106 for the terminal devices. As another example, the hub 114 may be a controller that sends commands or instructions to one or more actuators in the terminal devices. Commands or instructions may be received from the terminal devices, network nodes 110, or by executable code, script, process, or other instructions in the hub 114. As another example, the hub 114 may be a data collector that acts as temporary storage for terminal device data and, insome embodiments, may perform analysis or other processing of the data. As another example, the hub 114 may be a content source. For example, for a terminal device that is a VR headset, display, loudspeaker or other media delivery device, the hub 114 may retrieve VR assets, video, audio, or other media or data related to sensory information via a network node, which the hub 114 then provides to the terminal device either directly, after performing local processing, and / or after adding additional local content. In still another example, the hub 114 acts as a proxy server or orchestrator for the terminal devices, in particular in if one or more of the terminal devices are low energy loT devices.

[0063] The hub 114 may have a constant / persistent or intermittent connection to the network node 110b. The hub 114 may also allow for a different communication scheme and / or schedule between the hub 114 and terminal devices (e.g., terminal device 112c and / or 112d), and between the hub 114 and the core network 106. In other examples, the hub 114 is connected to the core network 106 and / or one or more terminal devices via a wired connection. Moreover, the hub 114 may be configured to connect to an M2M service provider over the access network 104 and / or to another terminal device over a direct connection. In some scenarios, terminal devices may establish a wireless connection with the network nodes 110 while still connected via the hub 114 via a wired or wireless connection. In some embodiments, the hub 114 may be a dedicated hub - that is, a hub whose primary function is to route communications to / from the terminal devices from / to the network node 110b. In other embodiments, the hub 114 may be a non-dedicated hub - that is, a device which is capable of operating to route communications between the terminal devices and network node 110b, but which is additionally capable of operating as a communication start and / or end point for certain data channels.

[0064] As described above, when observing some poor performance on cells, network engineers need to manually analyze which type of issue the cell is suffering and take some operations accordingly to improve the performance of the cell. Although there are some solutions based on single serving cell or neighbor cells or target nodes in a graph, an automated and efficient solution for network optimization is still needed.

[0065] Example embodiments of the present disclosure provide an improved solution for network optimization. In this solution, a subgraph for a serving cell of a communication network is generated based on determining that the serving cell is an issue cell. The subgraph comprises nodes and edges connecting the nodes, the nodes represent the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicates a relationship between two cells represented by the two nodes. Based on feature information of the nodes and the edges in the subgraph, the subgraph is classified into at least one of a plurality of network optimization action types. A network optimization action type indicates at least onenetwork optimization action to be applied to alleviate an issue in the serving cell. For each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell are determined. For each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell is determined. In summary, the solution contains three steps: 1) defining the type of issue, e.g., the at least one network optimization action type; 2) finding the target cell(s) from the serving and neighbor cells which contribute to each issue; 3) reconfiguring the settings of the target cell(s) by performing network optimization.

[0066] In this way, it is much easier to connect to the practice of network optimizing, and the optimization recommendations can be provided accordingly. The whole process can be automated to reduce manual work, increase efficiency, accuracy, and reliability. Furthermore, as the poor performance of an issue cell may be due to both serving cell itself and its neighbors, the subgraph is constructed, with the serving node representing the serving cell; neighbor nodes representing the neighbor cells. The optimization action type is assigned to subgraph instead of considering the optimization action type of a single cell. The advantage is the interaction between cells and neighbors is taken into consideration which is closer to the practice of network operation.

[0067] Embodiments of the present disclosure will be described in detail below with reference to the accompanying drawings.

[0068] Reference is now made to FIG. 2, which illustrates a flowchart of a process 200 for network optimization in accordance with some embodiments of the present disclosure. The process 200 may be implemented by the network optimization device 120 in FIG. 1. The following will be described with reference to FIG. 1.

[0069] At block 210, the network optimization device 120 performs subgraph extraction. Specifically, the network optimization device 120 may generate a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell. The term "serving cell" refers to a current cell that is actively handling communication with terminal devices and is detected as the issue cell. A serving cell may be identified or detected as an issue cell by various means, an example of which can be found in the International Patent Application published as WO2019233635A1.

[0070] The subgraph may comprise nodes and edges connecting the nodes. The nodes may represent the serving cell and at least one neighbor cell of the serving cell, respectively. An edge between two nodes may indicate a relationship between two cells represented by the two nodes.

[0071] In some example embodiments, the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

[0072] As the structure of the whole wireless network can be expressed as a graph, the nodes of the graph may represent the cells, and the edges may represent the relationship between cells. When a cell with issue(s) has been detected in the upstream tasks, a subgraph based on the issue cell and its associate direct neighbor relation cells can be extracted from the whole network. In such subgraph, the serving cell (i.e., the issue cell) is defined as a serving node (referred to as “S- node”), and the high-related neighbor cells are defined as neighbor nodes (referred to as “N-node”). In the following, the terms “serving node”, “serving cell” and “S-node” are used interchangeably, and the terms “neighbor node”, “neighbor cell” and “N-node” are used interchangeably.

[0073] In some example embodiments, the subgraph for the serving cell may be extracted from a graph corresponding to the communication network. The graph may comprise a set of nodes in the communication network and edges connecting the set of nodes. The set of nodes may represent respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

[0074] FIG. 3 illustrates an example of different stages at the process of network optimization in accordance with some further embodiments of the present disclosure. As shown in FIG. 3, there is a graph 310 which includes multiple nodes and edges therebetween. When determining that a serving cell is an issue cell (which is represented as S in FIG. 3), the network optimization device 120 may generate a subgraph 312 for the serving cell which includes some nodes corresponding to the serving cell and its neighbor cells (which are represented as N in FIG. 3). There are edges (represented by solid lines) between these nodes (represented by circles), to indicate the relationship between the connected nodes.

[0075] At block 220, the network optimization device 120 may perform subgraph classification. Specifically, the network optimization device 120 may classify, based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types. A network optimization action type may indicate at least one network optimization action or a network optimization solution to be applied to alleviate an issue in the serving cell.

[0076] As shown in FIG. 3, the serving cell is represented as a S-node (i.e., a first node with the first node type), and a neighbor cell is represented as a N-node (i.e., a second node with the second node type). When determining that the serving cell is an issue cell, may generate a subgraph 312 which includes the S-node corresponding to the serving cell and multiple N-nodes corresponding to the neighbor cells around this serving sell. There are edges between the S-node and the N-nodes.

[0077] As shown in FIG. 3, the S-node may have feature information, e.g., [si, s2, . . .], and a N-node may have feature information, such as [nl, n2, . . .]. Different nodes may have the sameor different types of feature information, which are not limited in the present disclosure. An edge may have feature information, such as [rel 1 , rel2, . . .].

[0078] In some example embodiments, the feature information of a node in the subgraph may indicate, but is not limited to, key performance indicator (KPI) information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node. The KPI information of a cell may include, but is not limited to, intra frequency HO failure ratio, inter frequency HO failure ratio, uplink / downlink user throughput, uplink / downlink cell throughput, and the like. In some cases, considering different impact of the S-node and N-node on the issue, the feature information of S-node and N-node may be different, which makes the subgraph heterogeneous. That is, such subgraph with different set of node features may be referred to as “heterogeneous subgraph” as described below and the network optimization recommendation solution disclosed herein is able to handle such heterogeneous subgraph.

[0079] In some example embodiments, the feature information of an edge between two nodes in the subgraph indicates a strength of the relationship between two cells represented by the two nodes. The edges between nodes may represent the relationship between S-node and N-nodes. In some cases, the feature information of an edge may indicate the strength of the relationship. For example, handover (HO) counts, and radio signal footprint overlap ratio are the edge attributes for handover issues.

[0080] In some example embodiments, the nodes in the subgraph may comprise a first node with a first node type and at least one second node with a second node type. Herein, the first node with a first node type may represent the serving cell and the first node type may indicate that the serving cell is an issue cell. The second node with the second node type may represent the neighbor cell, and the second node type may indicate a neighbor cell of the issue cell. A type of feature information of the first node is different from a type of feature information of the at least one second node.

[0081] In FIG. 3, based on feature information of the nodes and the edges in the subgraph 312, the network optimization device 120 may classify the subgraph 312 into a network optimization action type A 320 and a network optimization action type B 322. The network optimization action type A may indicate one network optimization action to be applied to alleviate an issue in the serving cell and the network optimization action type B may indicate another network optimization action to be applied to alleviate an issue in the serving cell. The issues to be addressed or alleviated through the network optimization actions of the types A and B may be the same or different.

[0082] At block 230, the network optimization device 120 may perform target node identification. Specifically, the network optimization device 120 may determine for each of the at least one network optimization action type, based on feature information of the nodes and theedges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell. A node representing a target cell in the subgraph may thus be referred to as a target node. In the following, the terms “target node” and “target cell” are used interchangeably.

[0083] For example, in FIG. 3, for the network optimization action type A 320, the network optimization device 120 may determine the node 330 (i.e., S-node) as the target node based on feature information of the nodes and the edges in the subgraph 312. For the network optimization action type B 322, the network optimization device 120 may determine the nodes 332 and 334 (i.e., N-nodes) as the target nodes based on feature information of the nodes and the edges in the subgraph 312.

[0084] At block 240, the network optimization device 120 may determine, for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell determined for the corresponding network optimization action type.

[0085] In this way, the subgraph extraction, the subgraph classification and the target node identification may be realized, thereby providing optimization recommendations to alleviate the issue in the serving cell based on the target nodes.

[0086] As the wireless network structure can be expressed as graphs, from which the issue cells are defined as serving node (S-node) and its high-related neighbors may be defined as neighbor nodes (N-node), and there generates a subgraph from the whole network with two node types containing different features. Within the scope of the issue, both adjusting operations (e.g., “increase”, “decrease”) and corresponding parameters are defined as multiple types of network optimization actions which are determined by the issue node and impact from its neighbor nodes. It means graph classification can be implemented on the heterogeneous graph with multi-label of the network optimization action types.

[0087] In some example embodiments, the network optimization device 120 may perform a subgraph classification on the subgraph using a GNN. The network optimization device 120 may obtain an output of the GNN. The output of the GNN at least comprises a plurality of labels corresponding to the plurality of network optimization action types, and each of the plurality of labels may indicate a predicted probability of the subgraph being classified into a corresponding network optimization action type. Further, the network optimization device 120 may determine the at least one network optimization action type for the subgraph based on the output of the GNN.

[0088] After the data of the serving cell and its neighbor cells being prepared, a GNN model may be used to perform subgraph classification and target nodes identification. In model training, the GNN model may receive heterogeneous subgraph as input, and train and predict it as a multilabel classification model. After the prediction, the GNN model may generate the contributing nodes (i.e., target nodes) to predict for each network optimization action type.

[0089] As the poor performance of an issue cell may be due to both the cell itself and its neighbors, the network optimization action type may be assigned to the subgraph instead of single serving node. Then the network optimization action type is defined as the label of subgraph. As there may be more than one network optimization action of a subgraph, the label of a subgraph may be various. When the labeled training dataset are provided by domain experts or assisted by clustering method, the problem can be formulated as subgraph classification with heterogeneous input feature information and multi-label output.

[0090] FIG. 4 illustrates a schematic diagram of training and application stages of GNN-based network optimization in accordance with some embodiments of the present disclosure. As shown in FIG. 4, in a GNN model training phase, a GNN 410 (also referred to as GNN model) may receive a sample subgraph 402 as input and output a type prediction 420. The sample subgraph 402 may include heterogeneous input feature information of the associated nodes, and the type prediction 420 may include multi-label output prediction, for example, prediction for each network optimization action type, such as type A, type B, type C,. . ., type X.

[0091] After the training of the GNN 410 is completed, in a GNN model application phase, the trained GNN 410 may be provided to perform network optimization. The GNN 410 may receive the generated subgraph 312 as its input and process the subgraph 312. The GNN 410 may provide the multi-label output for the subgraph 312, which indicates that the subgraph 312 is classified into the network optimization action type A and the network optimization action type B. Although only two network optimization action types are illustrated in FIG. 4, the trained GNN 410 may output classification probabilities and specific target node(s) for all network optimization action types, and one or more network optimization types may be determined for an input subgraph based on the output classification probabilities.

[0092] FIG. 5 illustrates an example 500 of multi-label prediction by the GNN in accordance with some further embodiments of the present disclosure. In conjunction with FIG. 4, the GNN 410 may output X classification probabilities for X network optimization action types, where the network optimization action types A and B have higher probabilities, the network optimization action types from D to X have lower probabilities. In this case, the network optimization action types A and B are determined as the network optimization action types for the subgraph 312.

[0093] In some example embodiments, the GNN may comprise a plurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph. In some example embodiments, considering the feature information of the S-node, N-nodes, and edges in the subgraph 312, the GNN may at least comprise one or more of: a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing theat least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, the intermediate feature representation being determined based on the feature information of the first node and the at least one second node. In other word, the intermediate feature representation may be used in the GNN for subgraph classification.

[0094] FIG. 6 illustrates a block diagram of an example model structure of the GNN in accordance with some embodiments of the present disclosure. The GNN 410 may include a heterogeneous layer and a homogeneous layer to perform subgraph classification with heterogeneous input features and realize multi-label output.

[0095] The heterogeneous layer may further include a dense layer 610 (i.e., first processing layer) for receiving feature information of the serving cell as its input, and a dense layer 612 (i.e., second processing layer) for receiving feature information of the neighbor cells as its input. For example, the dense layer 610 is configured to receive feature information 602 of the serving cell in a dataset consisting of a multi-dimensional (e.g., 11 -dimensional) vector of the S-node. For example, the dense layer 612 is configured to receive feature information 604 of the neighbor cells in a dataset consisting of a multi-dimensional (e.g., 3 -dimensional) vector of the N-nodes. The heterogeneous layer may utilize a number of output units (e.g., 64 output units) to align the dimension of feature information of the nodes and output an intermediate feature representation. The intermediate feature representation may be determined based on the feature information of the S-node and the N-node. It would be appreciated that there is no limitation on the dimension of the vectors and the number of units in each layer of the GNN, and those can be designed as need.

[0096] Basic GNN is to input the node features with same dimension for all nodes. However, the dimensions of node features for S-node and N-node applied in the present disclosure are different. To handle the different type of nodes, two independent layers with different dimensions are built to align the dimensions in heterogeneous layer. The general graph structural data with aligned node features may be obtained, and the homogeneous layer with dense layer and convolution layer may be built as well. Herein it needs to consider feature information of edges. Although the feature information of edges are often multiple dimensions in the wireless network, a basic convolution layer, such as Graph Convolutional Network (GCN), can only handle an edge weight with 1 dimension. To handle multi-dimensional feature information of edges, a convolution layer may be employed, such as Edge-conditioned convolution (ECC) layer.

[0097] Thus, referring to FIG. 6, the homogeneous layer may include a dense layer 620 with dropout, an Edge-Conditioned Convolution (ECC) layer 622 and a dense layer 624 with dropout. In the homogeneous layer, the dropout layers are added after the dense layers with activation. The dense layer 620 with dropout may receive the intermediate feature representation from theheterogeneous layer, and produce a further intermediate feature representation. To avoid confusion by message passing about the contribution of nodes to the classification (e.g., Grad-CAM), one ECC layer is employed. The ECC layer 622 may receive feature information 606 of edges in a dataset consisting of a multi-dimensional vector of the edges and the intermediate feature representation of the feature information of the S-node and N-nodes from the dense layer 620. As an example, the feature information 606 of edges is illustrated as a 2-dimensional vector although other dimension is also applicable, depending on the requirement on feature information of an edge.

[0098] Transforming of feature information in ECC can be defined as follows:where x£denotes features of a certain node z, 0 denotes edge-specific weights calculated by h0, h0denotes a neural network to transform feature information of edges, JV'(Z) denotes adjacent node for node tand e£j- denotes features of a certain edge between nodes i and j.

[0099] After feature information transformation, the GNN 410 may aggregate all nodes in a subgraph into one graph feature to classify it as graph classification using Global Average Pooling (GAP). Finally, to classify a subgraph for multi-label, the model uses dense layer with sigmoid- related activation function. The dimension of this layer is same as the number of classes exist in the training dataset.

[0100] Thus, referring to FIG. 6, the GNN 410 may further include a GAP layer 626 and a dense layer 628, where the GAP layer 626 is used to aggregate divided feature information of the nodes and edges into one feature and the dense layer 628 is used to classify subgraph, e.g., through sigmoid. That is, the GNN 410 may aggregate all nodes and edges in a subgraph into one graph feature to classify it as graph classification using GAP. Finally, the GNN 410 may provide a multilabel output 630 with sigmoid.

[0101] Then, the subgraph classification model, i.e., the GNN 410 will perform a multi-label prediction for each unseen issue cell and the output is optimization action(s). An example is shown in Table 1.Table 1 : Subgraph classification output

[0102] In some example embodiments, data may be collected from nation-wide live network for a period (e.g., several weeks), then may be used clustering method for subgraphs to form the above datasets. In this way, it may assist domain expert’s annotation which make it possible to get thousands of training data finally. Alternatively, the number of N-nodes, and the number of labels and their combinations are varied.

[0103] The above has described how to determine and classify the subgraph according to the example embodiments of the present disclosure. After that, the at least one target cell for each of the at least one network optimization action type needs to be determined, that is, the target nodes corresponding to the target cells should be determined.

[0104] In some example embodiments, for a given network optimization action type among the at least one network optimization action type, an intermediate feature representation of the feature information of the nodes and the edges in the subgraph may be extracted from the GNN. The feature information is provided as input to the GNN for processing. The classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation. Respective contribution scores of the nodes in the subgraph to the given network optimization action type are determined based on the intermediate feature representation. A contribution score of a node indicates a contribution of a cell represented by the node to the given network optimization action type. Then, at least one target cell for the given network optimization action type is determined based on the respective contribution scores of the nodes.

[0105] In some example embodiments, the GNN comprising a plurality of processing layers, and the intermediate feature representation is extracted from an output of any processing layer of the GNN.

[0106] In some example embodiments, Grad-CAM is one of the methods which can interpret neural network model using its gradient. In image classification using a Convolutional Neural Network (CNN), the gradient is calculated for each filter in a CNN layer and averaged for eachfilter to represent the importance of the filter. Then, from the layer in which the weights are obtained, the actual output values are multiplied by the corresponding weights. Finally, a Grad- CAM score is given as a contribution score by applying the min-max normalization. In graph classification, it needs to calculate the importance of each feature instead of the importance of each filter.

[0107] In some example embodiments, to determine respective contribution scores of the nodes in the subgraph to the given network optimization action type, respective weights of the nodes in the subgraph with respect to the given network optimization action type may be determined based on the intermediate feature representation. For example, the class weights for a class c at layer I and for feature k are calculated by the following equation:where F^n(i.e., intermediate feature representation) represents an output from a unit k at a layer I for a node n in the GNN; K represents the number of node features and k =C represents the number of classes (i.e., the network optimization types) and c = 1, . . . , C; N represents the number of nodes in the subgraph and n = 1, . . . , N; ycindicates a classification probability for the class c, i.e., the classification probability of a network optimization type.

[0108] FIG. 7 illustrates a schematic diagram of an example of contribution score determination in accordance with some embodiments of the present disclosure. As shown in FIG. 7, a subgraph may include N node, such as node 1, . . ., node N. F^n, as an intermediate feature representation, represents an output from a unit k at a layer I for a node n in the GNN. Equation (2) represents the gradient of the GNN model from F^nto class c.

[0109] In some example embodiments, a transformation operation on a weighted intermediate feature representation may be applied. For example, a ReLU function is applied. The weighted intermediate feature representation is obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation. Further, respective contribution scores of the nodes to the given network optimization action type may be determined based on the transformed intermediate feature representation, e.g., by applying min-max normalization on the results of the ReLU function. As an example, using above weights, the contribution scores of respective nodes for a given network optimization type are calculated as follows:where I is needed to be determined to one layer of the GNN. In some example embodiments, with reference to FIG. 6, a dense layer before the GAP layer may be set to calculate the score. Finally, the contribution scores for each network optimization type can be obtained through min-max normalization for each subgraph. To determine the target nodes, it needs to define the contribution threshold for contribution scores.

[0110] In some example embodiments, the transformed intermediate feature representation may comprise non-normalized contribution scores of the nodes to the given network optimization action type. To determine respective contribution scores of the nodes based on the transformed intermediate feature representation, the non-normalized contribution scores of the nodes may be adjusted based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and a normalization operation may be applied on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.[oni] FIG. 8 illustrates a flowchart of general flow 810 for contribution score determination in accordance with some embodiments of the present disclosure.

[0112] As shown in FIG. 8, non-normalized contribution scores are shown in row 812, for example, a non-normalized contribution score of S-node is 2.5, a non-normalized contribution score of one N-node is -1 and a non-normalized contribution score of another N-node is 5. After applying activation (e.g., ReLU), as shown in row 814, the contribution score of S-node is 2.5, the scores of the above N-nodes are 0 and 5, respectively. After normalization, as shown in row 816, the contribution score of S-node is 0.5 and the scores of the above N-nodes are 0 and 1, respectively.

[0113] In some example embodiments, adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell. If it is determined that the given network optimization action type is associated with the at least one neighbor cell, a non-normalized contribution score of a first node representing the serving cell may be replaced with a score of zero. If it is determined that the given network optimization action type is associated with the serving cell, at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell may be replaced with a score of zero.

[0114] In the heterogeneous network, the node type (i.e., S-node or N-node) to be applied depends on a network solution. In order to limit the node type that should get a higher score depending on the solution, a method may be proposed to calculate the final contribution score by setting the score of the non-applicable nodes as zero when calculating it. That is, after applying ReLU in Equation (3) and before applying min-max normalization, the score of the nodes which does not have corresponding node type is replaced by 0. In more detail, the score of S-node isreplaced by 0 if a solution is for N-node, and the score of N-nodes are replaced by 0 if a solution is for S-node. This allows for avoiding recommendations to nodes with inappropriate node type. The example of the procedure to recommend solution for N-node is shown in FIG. 8, which further illustrates a flowchart of node type restricted flow 820 for contribution score determination in accordance with some embodiments of the present disclosure.

[0115] As shown in FIG. 8, in this case, it is determined that the given network optimization action type is associated with the neighbor cell, and a non-normalized contribution score of S-node may be replaced with a score of zero. Specifically, non-normalized contribution scores are shown in row 822, for example, a non-normalized contribution score of S-node is 2.5, a non-normalized contribution score of one N-node is -1 and a non-normalized contribution score of another N-node is 5. After applying activation (e.g., ReLU), as shown in row 824, the contribution score of S-node is 2.5, the scores of the above N-nodes are 0 and 5, respectively. For limiting the node type, as shown in row 826, the contribution score of S-node is set to zero. After normalization, as shown in row 828, the contribution score of S-node is 0 and the scores of the above N-nodes are 0 and 1, respectively.

[0116] Following the network optimization action prediction, the target nodes identification module works on the calculation of the contribution of each node for the classification and get target node(cell) associated to the specific optimization action. Then the final network optimization recommendation output will be as shown in Table 2.Table 2: Subgraph classification and target nodes identification output

[0117] According to the above embodiments, targets of solution recommendation are prediction of network optimization action type and identification of target node for each optimization action type. Network optimization action types are predicted by GNN classification model and target node identification is done by determining the contribution scores of the nodes in the subgraph. To determine the target nodes, it needs to define the contribution threshold for each network optimization type. Although it can be set to a value (e.g., 0.5) simply, it might be different by network optimization type.

[0118] In some example embodiments, for determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes, the respective contribution scores of the nodes may be compared with a contribution threshold for the given network optimization action type. At least one of the nodes may be selected with a contribution score exceeding the contribution threshold. Further, at least one cell represented by the at least one selected node in the subgraph may be determined as the at least one target cell for the given network optimization action type.

[0119] To define the contribution thresholds dynamically, some subgraphs may be randomly sampled from a training dataset of the GNN and ground-truth target cells may be created for them. If it is done, an appropriate threshold for a network optimization action type may be defined by changing thresholds and checking the result of target cells identification for each network optimization action type.

[0120] Specially, in some example embodiments, in order to determine a suitable contribution threshold for each given network optimization action type, a set of sample subgraphs classified into the given network optimization action type may be sampled from a training dataset for the GNN. A set of ground-truth target cells in the set of sample subgraphs are determined for the given network optimization action type. Those ground-truth target cells are known to contribute to the given network optimization action type in the sample subgraphs. Then it is attempted to determine, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given networkoptimization action type with a plurality of candidate contribution thresholds. The plurality of sets of candidate target cells are compared with the set of ground-truth target cells, to determine respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells. Then the contribution threshold for the given network optimization action type is selected from the plurality of candidate contribution thresholds based on the respective matching ratios of the plurality of sets of candidate target cells. For example, the candidate contribution threshold with the highest matching ratio may be selected for the given network optimization action type.

[0121] To evaluate target nodes identification by the contribution scores, an exact match ratio (EMR) may be used to indicate the percentage of subgraphs where all target nodes are completely and correctly detected. Since tuning configuration is very sensitive in wireless network optimization, the target nodes identification may be evaluated by using EMR. EMR can be calculated as following:where J is the function indicating / (true) = 1 and / (false) = 0, represents a ground-truth label of i-th subgraph, represents label predicted by the GNN for i-th subgraph, and m represents the number of subgraphs contained in dataset.

[0122] When determining the contribution threshold, the optimal value may differ from network optimization type to type. To determine it for each network optimization type, sample subgraphs with ground-truth target nodes and candidate thresholds [0,0.1,0.2,. . .,0.9,1] may be prepared, and the threshold with the highest EMR is the final threshold. Herein, the sample subgraphs to be assigned the ground-truth target nodes used to determine the threshold are limited to those whose scores have values other than 0 and 1. Since contribution scores are eventually normalized from 0 to 1 by min-max normalization, all nodes in a subgraph often have only 0 or 1. If the ground-truth target nodes are given for such subgraphs, it doesn’t help to know the boundaries of ambiguous contribution scores to determine the target nodes. For this reason, to prepare dataset with the ground-truth target nodes efficiently, the sample subgraphs to be selected are limited by the calculated contribution score.

[0123] Table 3 shows the EMR of the entire dataset obtained by the percentage of data annotated with ground-truth target nodes and the contribution threshold that was the best selected by it for the class with target node type is N-node.Table 3: Dynamic determined thresholds and EMRwhere th denotes the contribution threshold selected by the sample dataset with ground-truth target nodes. From this table, with only 1% of the data with ground-truth target nodes, the high EMR for target nodes identification can be obtained.

[0124] An example of multi-label output label will be provided as follows. Table 4 shows that the relation between label and KPIs.Table 4: Description of the relation between Label and KPI

[0125] One example of prediction and target nodes identification for a subgraph is shown. The raw feature information for S-node, N-node and Edge are shown in Tables 5, 6 and 7, respectively.Table 5: S-node FeaturesTable 6: N-node FeaturesTable 7: Edge Features

[0126] After prediction by the GNN, according to the label definition, the most explicit labels for the example case are A2 and D3. In target nodes identification, the contribution scores are calculated (e.g., through Grad-CAM) for label A2 and D3.

[0127] For label A2, the correspond features of nodes and edges are:• S-node: featnd s Ol represents serving cell’s intra frequency overlap condition.• N-node: featnd n Ol represents neighbor cell’s tilt setting, and featnd_n_02 represents neighbor cell’s overshooting condition.• Edge: feated Ol represents the correlation between serving cell and neighbor cell by measurement report; and feated_02 represents the correlation between serving cell and neighbor cell by HO statistic.

[0128] For label D3, the correspond features of nodes and edges are:• S-node: featnd s Ol represents serving cell’s intra frequency overlap condition; featnd_s_02 represents the ratio of inter frequency HO failure in serving cell; and featnd_s_03 represents serving cell’s event A3 TTT.

[0129] Based on Tables 5, 6, and 7, regarding label A2, S-node nl l l l shows high overlap, even though N-node nl 115 and nl 116 are overshooting and with room for down tilt, only nl 115 has high correlation with S-node represented by the Edge attributes. So, this subgraph is predicted as label A2 correctly, and finally N-node nl 115 is the ground-truth target node which was detected by calculating contribution sores correctly. Regarding label D3, S-node nl l l l shows high overlap,high intra frequency HO failure ratio as well as large event A3 TTT. So, this subgraph is predicted as label D3 correctly, and S-node nl 111 is a target node determined by target node type.

[0130] In the above discussion, a method according to the embodiments of the present disclosure is proposed based on GNN and explanation technique of contribution scores to implement end-to-end subgraph network optimization action type classification and target nodes identification. And the output of the method provides optimization recommendations which include network optimization action types (adjusting operations and corresponding parameters) and the target nodes to implement these actions.

[0131] The solution in accordance with the embodiments of the present disclosure has various advantages.

[0132] Comparing with existing solutions, this method can give the optimization action and the target nodes to be reconfigured in one machine leaning module. When the action types and target nodes need to be reconfigured, it is much easier to connect to the practice of network optimizing, and the optimization recommendations can be provided accordingly. The whole process can be automated to reduce manual work, increase efficiency, accuracy and reliability.

[0133] As the poor performance of an issue cell may be due to both serving cell itself and its neighbors, the subgraph is constructed, with the serving node representing the serving cell; neighbor nodes representing the neighbor cells. And the optimization action type is assigned to subgraph instead of considering the optimization action type of a single cell. The advantage is the interaction between cells and neighbors is taken into consideration which is closer to the practice of network operation.

[0134] As the subgraph is constructed, the GNN method can be implemented to the subgraph classification. However, considering the input and output in practice, the features can be totally different for different types of nodes (S-node features and N-node features) which means the input features are heterogeneous. Additionally, the network optimization action type of a subgraph varies, which means the label of one subgraph could be more than one. Therefore, the GNN model proposed here can handle metrics that can be different for both serving cell and its neighboring cells and multi-label graph classification.

[0135] Once a subgraph is classified, the target node within the subgraph is identified to be reconfigured by calculating contribution scores, the nodes with higher contribution scores are defined as target nodes. The contribution scores can also decompose the impact of nodes to different action types.

[0136] Based on GNN models and calculating contribution scores of the nodes directly on GNN, it does not need extra training process for explainer. The classification labels and target nodes are given in one recommendation output.

[0137] As the contribution score aims to indicate the importance or contribution of each node to an optimization action type, it may highlight either serving node or neighbor nodes that are needed to be optimized. If the highlighted node type is neighbor nodes, the specific target node(s) out of all neighboring nodes is then identified. If the node type is serving cell, then the target node is the issue node. The target node type (either serving node or neighbor node(s)) is determined by the subgraph label; based on the radio domain expertise, thus the optimization target node is constrained by nodes type.

[0138] FIG. 9 illustrates a block diagram of an example cloud environment 900 in accordance with some embodiments of the present disclosure. The example cloud environment 900 may be containerized and deployed as virtual function on cloud or server host that support container environment. The whole solution includes the following components.

[0139] The example cloud environment 900 may include a continuous integration / continuous delivery (CI / CD) module 910, a data fetcher module 920, an extract / transform / load pipeline (ETL pipeline) module 930, a machine learning operations (MLOPs) module 940 and a monitoring module 950. The CI / CD module 910 is configured to deploy the whole solution end- to-end. The data fetcher module 920 is configured to fetch files from data source. The ETL pipeline module 930 is configured to process data. The MLOPs module 940 is configured to open a model service and get the prediction by sending requests. The MLOPs module 940 may further include a model service module 942 and a model request module 944. The monitoring module 950 may include a visualized dashboard to monitor all the process.

[0140] FIG. 10 is a flowchart of a computer-implemented method 1000 for network optimization in accordance with some embodiments of the present disclosure. The method 1000 may be implemented in any suitable apparatus or system with the computing capability, to support the network optimization in a communication network. In some embodiments, the method 1000 may be implemented at the cloud environment 900 of FIG. 9. In some embodiments, the method 1000 may be implemented at a network node in the access network or the core network of the communication system 100 of FIG. 1, such as the network optimization device 120. For the purpose of discussion, The method 1000 is discussed from the perspective of the network optimization device 120 .

[0141] At block 1010, the network optimization device 120 generates a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes.

[0142] At block 1020, the network optimization device 120 classifies, based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell.

[0143] At block 1030, the network optimization device 120 determines for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell.

[0144] At block 1030, the network optimization device 120 determines for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell.

[0145] In some embodiments, the nodes in the subgraph comprise: a first node with a first node type representing the serving cell, the first node type indicating that the serving cell is an issue cell, and at least one second node with a second node type representing the at least one neighbor cell, the second node type indicating a neighbor cell of the issue cell; and wherein a type of feature information of the first node is different from a type of feature information of the at least one second node.

[0146] In some embodiments, the network optimization device 120 extracts, based on the serving cell, the subgraph from a graph corresponding to the communication network, the graph comprising a set of nodes and edges connecting the set of nodes, the set of nodes representing respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

[0147] In some embodiments, the network optimization device 120 performs a subgraph classification on the subgraph using a graph neural network (GNN), the GNN comprising a plurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph; obtains an output of the GNN, the output of the GNN at least comprising a plurality of labels corresponding to the plurality of network optimization action types, each of the plurality of labels indicating a predicted probability of the subgraph being classified into a corresponding network optimization action type; and determines the at least one network optimization action type for the subgraph based on the output of the GNN.

[0148] In some embodiments, the GNN at least comprises one or more of a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing the at least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, theintermediate feature representation being determined based on the feature information of the first node and the at least one second node.

[0149] In some embodiments, the network optimization device 120, for a given network optimization action type among the at least one network optimization action type, extracts an intermediate feature representation of the feature information of the nodes and the edges in the subgraph from a GNN, wherein the feature information is provided as input to the GNN for processing, and the classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation; determines respective contribution scores of the nodes in the subgraph to the given network optimization action type based on the intermediate feature representation, a contribution score of a node indicating a contribution of a cell represented by the node to the given network optimization action type; and determines at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes.

[0150] In some embodiments, the GNN comprises a plurality of processing layers, and the intermediate feature representation is extracted from an output of a processing layer of the GNN.

[0151] In some embodiments, the network optimization device 120 determines, based on the intermediate feature representation, respective weights of the nodes in the subgraph with respect to the given network optimization action type; applies a transformation operation on a weighted intermediate feature representation, the weighted intermediate feature representation being obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation; and determines respective contribution scores of the nodes to the given network optimization action type based on the transformed intermediate feature representation.

[0152] In some embodiments, the transformed intermediate feature representation comprises non-normalized contribution scores of the nodes to the given network optimization action type, and the network optimization device 120 adjusts the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and applies a normalization operation on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.

[0153] In some embodiments, the network optimization device 120, in accordance with a determination that the given network optimization action type is associated with the at least one neighbor cell, replaces a non-normalized contribution score of a first node representing the servingcell with a score of zero; and in accordance with a determination that the given network optimization action type is associated with the serving cell, replaces at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell with a score of zero.

[0154] In some embodiments, the network optimization device 120 compares the respective contribution scores of the nodes with a contribution threshold for the given network optimization action type; selects at least one of the nodes with a contribution score exceeding the contribution threshold; and determines at least one cell represented by the at least one selected node in the subgraph as the at least one target cell for the given network optimization action type.

[0155] In some embodiments, the network optimization device 120 determines the contribution threshold for the given network optimization action type by: obtaining a set of sample subgraphs classified into the given network optimization action type, and a set of ground-truth target cells in the set of sample subgraphs for the given network optimization action type, wherein the set of sample subgraphs are sampled from a training dataset for the GNN; determining, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given network optimization action type with a plurality of candidate contribution thresholds; determining respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells; and selecting, from the plurality of candidate contribution thresholds, the contribution threshold for the given network optimization action type based on the respective matching ratios of the plurality of sets of candidate target cells.

[0156] In some embodiments, the feature information of a node in the subgraph indicates at least one of the following: key performance indicator information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node; and / or wherein the feature information of an edge between two nodes in the subgraph indicates a strength of the relationship between two cells represented by the two nodes.

[0157] In some embodiments, the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

[0158] FIG. 11 is a block diagram showing an apparatus 1100 suitable for use in practicing some embodiments of the disclosure. For example, any one of devices described above, including the terminal devices and the base station, may be implemented through the apparatus 1100. As shown, the apparatus 1100 may include a processor 1110, a memory 1120 that stores a program, and optionally a communication interface 1130 for communicating data with other external devices through wired and / or wireless communication.

[0159] The program includes program instructions that, when executed by the processor 1110, enable the apparatus 1100 to operate in accordance with the embodiments of the present disclosure, as discussed above. That is, the embodiments of the present disclosure may be implemented at least in part by computer software executable by the processor 1110, or by hardware, or by a combination of software and hardware.

[0160] The memory 1120 may be of any type suitable to the local technical environment and may be implemented using any suitable data storage technology, such as semiconductor-based memory devices, flash memories, magnetic memory devices and systems, optical memory devices and systems, fixed memories and removable memories. The processor 1110 may be of any type suitable to the local technical environment, and may include one or more of general-purpose computers, special purpose computers, microprocessors, digital signal processors (DSPs) and processors based on multi-core processor architectures, as non-limiting examples.

[0161] FIG. 12 is a block diagram of a host 1200, which may be an embodiment of the host 116 of FIG. 1, in accordance with various aspects described herein. As used herein, the host 1200 may be or comprise various combinations hardware and / or software, including a standalone server, a blade server, a cloud-implemented server, a distributed server, a virtual machine, container, or processing resources in a server farm. The host 1200 may provide one or more services to one or more UEs.

[0162] The host 1200 includes processing circuitry 1202 that is operatively coupled via a bus 1204 to an input / output interface 1206, a network interface 1208, a power source 1210, and a memory 1212. Other components may be included in other embodiments.

[0163] The memory 1212 may include one or more computer programs including one or more host application programs 1214 and data 1216, which may include user data, e.g., data generated by a UE for the host 1200 or data generated by the host 1200 for a UE. Embodiments of the host 1200 may utilize only a subset or all of the components shown. The host application programs 1214 may be implemented in a container-based architecture and may provide support for video codecs (e.g., Versatile Video Coding (VVC), High Efficiency Video Coding (HEVC), Advanced Video Coding (AVC), MPEG, VP9) and audio codecs (e.g., FLAC, Advanced Audio Coding (AAC), MPEG, G.711), including transcoding for multiple different classes, types, or implementations of UEs (e.g., handsets, desktop computers, wearable display systems, heads-up display systems). The host application programs 1214 may also provide for user authentication and licensing checks and may periodically report health, routes, and content availability to a central node, such as a device in or on the edge of a core network. Accordingly, the host 1200 may select and / or indicate a different host for over-the-top services for a UE. The host application programs 1214 may support various protocols, such as the HTTP Live Streaming (HLS) protocol, Real-TimeMessaging Protocol (RTMP), Real-Time Streaming Protocol (RTSP), Dynamic Adaptive Streaming over HTTP (MPEG-DASH), etc.

[0164] FIG. 13 is a block diagram illustrating a virtualization environment 1300 in which functions implemented by some embodiments may be virtualized. In the present context, virtualizing means creating virtual versions of apparatuses or devices which may include virtualizing hardware platforms, storage devices and networking resources. As used herein, virtualization can be applied to any device described herein, or components thereof, and relates to an implementation in which at least a portion of the functionality is implemented as one or more virtual components. Some or all of the functions described herein may be implemented as virtual components executed by one or more virtual machines (VMs) implemented in one or more virtual environments 1300 hosted by one or more of hardware nodes, such as a hardware computing device that operates as a network node, UE, core network node, or host. Further, in embodiments in which the virtual node does not require radio connectivity (e.g., a core network node or host), then the node may be entirely virtualized.

[0165] Applications 1302 (which may alternatively be called software instances, virtual appliances, network functions, virtual nodes, virtual network functions, etc.) are run in the virtualization environment 1300 to implement some of the features, functions, and / or benefits of some of the embodiments disclosed herein.

[0166] Hardware 1304 includes processing circuitry, memory that stores software and / or instructions executable by hardware processing circuitry, and / or other hardware devices as described herein, such as a network interface, input / output interface, and so forth. Software may be executed by the processing circuitry to instantiate one or more virtualization layers 1306 (also referred to as hypervisors or virtual machine monitors (VMMs)), provide VMs 1308A and 1308B (one or more of which may be generally referred to as VMs 1308), and / or perform any of the functions, features and / or benefits described in relation with some embodiments described herein. The virtualization layer 1306 may present a virtual operating platform that appears like networking hardware to the VMs 1308.

[0167] The VMs 1308 comprise virtual processing, virtual memory, virtual networking or interface and virtual storage, and may be run by a corresponding virtualization layer 1306. Different embodiments of the instance of a virtual appliance 1302 may be implemented on one or more of VMs 1308, and the implementations may be made in different ways. Virtualization of the hardware is in some contexts referred to as network function virtualization (NFV). NFV may be used to consolidate many network equipment types onto industry standard high volume server hardware, physical switches, and physical storage, which can be located in data centers, and customer premise equipment.

[0168] In the context of NFV, a VM 1308 may be a software implementation of a physical machine that runs programs as if they were executing on a physical, non-virtualized machine. Each of the VMs 1308, and that part of hardware 1304 that executes that VM, be it hardware dedicated to that VM and / or hardware shared by that VM with others of the VMs, forms separate virtual network elements. Still in the context of NFV, a virtual network function is responsible for handling specific network functions that run in one or more VMs 1308 on top of the hardware 1304 and corresponds to the application 1302.

[0169] Hardware 1304 may be implemented in a standalone network node with generic or specific components. Hardware 1304 may implement some functions via virtualization. Alternatively, hardware 1304 may be part of a larger cluster of hardware (e.g. such as in a data center or CPE) where many hardware nodes work together and are managed via management and orchestration 1310, which, among others, oversees lifecycle management of applications 1302. In some embodiments, hardware 1304 is coupled to one or more radio units that each include one or more transmitters and one or more receivers that may be coupled to one or more antennas. Radio units may communicate directly with other hardware nodes via one or more appropriate network interfaces and may be used in combination with the virtual components to provide a virtual node with radio capabilities, such as a radio access node or a base station. In some embodiments, some signaling can be provided with the use of a control system 1312 which may alternatively be used for communication between hardware nodes and radio units.

[0170] FIG. 14 shows a communication diagram of a host 1402 communicating via a network node 1404 with a UE 1406 over a partially wireless connection in accordance with some embodiments. Example implementations, in accordance with various embodiments, of the UE (such as a UE 112a of FIG. 1), network node (such as network node 110a of FIG. 1), and host (such as host 116 of FIG. 1 and / or host 1200 of FIG. 12) discussed in the preceding paragraphs will now be described with reference to FIG. 14.

[0171] Like host 1200, embodiments of host 1402 include hardware, such as a communication interface, processing circuitry, and memory. The host 1402 also includes software, which is stored in or accessible by the host 1402 and executable by the processing circuitry. The software includes a host application that may be operable to provide a service to a remote user, such as the UE 1406 connecting via an over-the-top (OTT) connection 1450 extending between the UE 1406 and host 1402. In providing the service to the remote user, a host application may provide user data which is transmitted using the OTT connection 1450.

[0172] The network node 1404 includes hardware enabling it to communicate with the host 1402 and UE 1406. The connection 1460 may be direct or pass through a core network (like core network 106 of FIG. 1) and / or one or more other intermediate networks, such as one or more public,private, or hosted networks. For example, an intermediate network may be a backbone network or the Internet.

[0173] The UE 1406 includes hardware and software, which is stored in or accessible by UE 1406 and executable by the UE’s processing circuitry. The software includes a client application, such as a web browser or operator-specific “app” that may be operable to provide a service to a human or non-human user via UE 1406 with the support of the host 1402. In the host 1402, an executing host application may communicate with the executing client application via the OTT connection 1450 terminating at the UE 1406 and host 1402. In providing the service to the user, the UE's client application may receive request data from the host's host application and provide user data in response to the request data. The OTT connection 1450 may transfer both the request data and the user data. The UE's client application may interact with the user to generate the user data that it provides to the host application through the OTT connection 1450.

[0174] The OTT connection 1450 may extend via a connection 1460 between the host 1402 and the network node 1404 and via a wireless connection 1470 between the network node 1404 and the UE 1406 to provide the connection between the host 1402 and the UE 1406. The connection 1460 and wireless connection 1470, over which the OTT connection 1450 may be provided, have been drawn abstractly to illustrate the communication between the host 1402 and the UE 1406 via the network node 1404, without explicit reference to any intermediary devices and the precise routing of messages via these devices.

[0175] As an example of transmitting data via the OTT connection 1450, in step 1408, the host 1402 provides user data, which may be performed by executing a host application. In some embodiments, the user data is associated with a particular human user interacting with the UE 1406. In other embodiments, the user data is associated with a UE 1406 that shares data with the host 1402 without explicit human interaction. In step 1410, the host 1402 initiates a transmission carrying the user data towards the UE 1406. The host 1402 may initiate the transmission responsive to a request transmitted by the UE 1406. The request may be caused by human interaction with the UE 1406 or by operation of the client application executing on the UE 1406. The transmission may pass via the network node 1404, in accordance with the teachings of the embodiments described throughout this disclosure. Accordingly, in step 1412, the network node 1404 transmits to the UE 1406 the user data that was carried in the transmission that the host 1402 initiated, in accordance with the teachings of the embodiments described throughout this disclosure. In step 1414, the UE 1406 receives the user data carried in the transmission, which may be performed by a client application executed on the UE 1406 associated with the host application executed by the host 1402.

[0176] In some examples, the UE 1406 executes a client application which provides user data to the host 1402. The user data may be provided in reaction or response to the data received from the host 1402. Accordingly, in step 1416, the UE 1406 may provide user data, which may be performed by executing the client application. In providing the user data, the client application may further consider user input received from the user via an input / output interface of the UE 1406. Regardless of the specific manner in which the user data was provided, the UE 1406 initiates, in step 1418, transmission of the user data towards the host 1402 via the network node 1404. In step 1420, in accordance with the teachings of the embodiments described throughout this disclosure, the network node 1404 receives user data from the UE 1406 and initiates transmission of the received user data towards the host 1402. In step 1422, the host 1402 receives the user data carried in the transmission initiated by the UE 1406.

[0177] One or more of the various embodiments improve the performance of OTT services provided to the UE 1406 using the OTT connection 1450, in which the wireless connection 1470 forms the last segment.

[0178] In an example scenario, factory status information may be collected and analyzed by the host 1402. As another example, the host 1402 may process audio and video data which may have been retrieved from a UE for use in creating maps. As another example, the host 1402 may collect and analyze real-time data to assist in controlling vehicle congestion (e.g., controlling traffic lights). As another example, the host 1402 may store surveillance video uploaded by a UE. As another example, the host 1402 may store or control access to media content such as video, audio, VR or AR which it can broadcast, multicast or unicast to UEs. As other examples, the host 1402 may be used for energy pricing, remote control of non-time critical electrical load to balance power generation needs, location services, presentation services (such as compiling diagrams etc. from data collected from remote devices), or any other function of collecting, retrieving, storing, analyzing and / or transmitting data.

[0179] In some examples, a measurement procedure may be provided for the purpose of monitoring data rate, latency and other factors on which the one or more embodiments improve. There may further be an optional network functionality for reconfiguring the OTT connection 1450 between the host 1402 and UE 1406, in response to variations in the measurement results. The measurement procedure and / or the network functionality for reconfiguring the OTT connection may be implemented in software and hardware of the host 1402 and / or UE 1406. In some embodiments, sensors (not shown) may be deployed in or in association with other devices through which the OTT connection 1450 passes; the sensors may participate in the measurement procedure by supplying values of the monitored quantities exemplified above, or supplying values of other physical quantities from which software may compute or estimate the monitored quantities. Thereconfiguring of the OTT connection 1450 may include message format, retransmission settings, preferred routing etc.; the reconfiguring need not directly alter the operation of the network node 1404. Such procedures and functionalities may be known and practiced in the art. In certain embodiments, measurements may involve proprietary UE signaling that facilitates measurements of throughput, propagation times, latency and the like, by the host 1402. The measurements may be implemented in that software causes messages to be transmitted, in particular empty or ‘dummy’ messages, using the OTT connection 1450 while monitoring propagation times, errors, etc.

[0180] Although the computing devices described herein (e.g., UEs, network nodes, hosts) may include the illustrated combination of hardware components, other embodiments may comprise computing devices with different combinations of components. It is to be understood that these computing devices may comprise any suitable combination of hardware and / or software needed to perform the tasks, features, functions and methods disclosed herein. Determining, calculating, obtaining or similar operations described herein may be performed by processing circuitry, which may process information by, for example, converting the obtained information into other information, comparing the obtained information or converted information to information stored in the network node, and / or performing one or more operations based on the obtained information or converted information, and as a result of said processing making a determination. Moreover, while components are depicted as single boxes located within a larger box, or nested within multiple boxes, in practice, computing devices may comprise multiple different physical components that make up a single illustrated component, and functionality may be partitioned between separate components. For example, a communication interface may be configured to include any of the components described herein, and / or the functionality of the components may be partitioned between the processing circuitry and the communication interface. In another example, non-computationally intensive functions of any of such components may be implemented in software or firmware and computationally intensive functions may be implemented in hardware.

[0181] In certain embodiments, some or all of the functionalities described herein may be provided by processing circuitry executing instructions stored on in memory, which in certain embodiments may be a computer program product in the form of a non-transitory computer- readable storage medium. In alternative embodiments, some or all of the functionalities may be provided by the processing circuitry without executing instructions stored on a separate or discrete device-readable storage medium, such as in a hard-wired manner. In any of those particular embodiments, whether executing instructions stored on a non-transitory computer-readable storage medium or not, the processing circuitry can be configured to perform the described functionality. The benefits provided by such functionality are not limited to the processingcircuitry alone or to other components of the computing device, but are enjoyed by the computing device as a whole, and / or by end users and a wireless network generally.

[0182] In some example embodiments, an apparatus capable of performing any of the method 1000 (for example, the network optimization device 120 in FIG. 1) may comprise means for performing the respective operations of the method 1000. The means may be implemented in any suitable form. For example, the means may be implemented in a circuitry or software module. The apparatus may be implemented as or included in the network optimization device 120.

[0183] In some embodiments, the apparatus comprises means for generating a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes; means for classifying, based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell; means for determining for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and means for determining for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell.

[0184] In some embodiments, the nodes in the subgraph comprise: a first node with a first node type representing the serving cell, the first node type indicating that the serving cell is an issue cell, and at least one second node with a second node type representing the at least one neighbor cell, the second node type indicating a neighbor cell of the issue cell; and wherein a type of feature information of the first node is different from a type of feature information of the at least one second node.

[0185] In some embodiments, the means for generating a subgraph for the serving cell comprises: means for extracting, based on the serving cell, the subgraph from a graph corresponding to the communication network, the graph comprising a set of nodes and edges connecting the set of nodes, the set of nodes representing respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

[0186] In some embodiments, the means for classifying the subgraph into at least one of a plurality of network optimization action types comprises: means for performing a subgraph classification on the subgraph using a graph neural network (GNN), the GNN comprising aplurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph; means for obtaining an output of the GNN, the output of the GNN at least comprising a plurality of labels corresponding to the plurality of network optimization action types, each of the plurality of labels indicating a predicted probability of the subgraph being classified into a corresponding network optimization action type; and means for determining the at least one network optimization action type for the subgraph based on the output of the GNN.

[0187] In some embodiments, the GNN at least comprises one or more of: a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing the at least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, the intermediate feature representation being determined based on the feature information of the first node and the at least one second node.

[0188] In some embodiments, the means for determining the at least one target cell for each of the at least one network optimization action type comprises: for a given network optimization action type among the at least one network optimization action type, means for extracting an intermediate feature representation of the feature information of the nodes and the edges in the subgraph from a GNN, wherein the feature information is provided as input to the GNN for processing, and the classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation; means for determining respective contribution scores of the nodes in the subgraph to the given network optimization action type based on the intermediate feature representation, a contribution score of a node indicating a contribution of a cell represented by the node to the given network optimization action type; and means for determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes.

[0189] In some embodiments, the GNN comprising a plurality of processing layers, and the intermediate feature representation is extracted from an output of a processing layer of the GNN.

[0190] In some embodiments, the means for determining respective contribution scores of the nodes in the subgraph to the given network optimization action type comprises: means for determining, based on the intermediate feature representation, respective weights of the nodes in the subgraph with respect to the given network optimization action type; means for applying a transformation operation on a weighted intermediate feature representation, the weighted intermediate feature representation being obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation; and means for determiningrespective contribution scores of the nodes to the given network optimization action type based on the transformed intermediate feature representation.

[0191] In some embodiments, the transformed intermediate feature representation comprises non-normalized contribution scores of the nodes to the given network optimization action type, and the means for determining respective contribution scores of the nodes based on the transformed intermediate feature representation comprises: means for adjusting the nonnormalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and means for applying a normalization operation on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.

[0192] In some embodiments, the means for adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell comprises: means for, in accordance with a determination that the given network optimization action type is associated with the at least one neighbor cell, replacing a non-normalized contribution score of a first node representing the serving cell with a score of zero; and means for, in accordance with a determination that the given network optimization action type is associated with the serving cell, replacing at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell with a score of zero.

[0193] In some embodiments, the means for determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes comprises: means for comparing the respective contribution scores of the nodes with a contribution threshold for the given network optimization action type; means for selecting at least one of the nodes with a contribution score exceeding the contribution threshold; and means for determining at least one cell represented by the at least one selected node in the subgraph as the at least one target cell for the given network optimization action type.

[0194] In some embodiments, the apparatus further comprises: means for determining the contribution threshold for the given network optimization action type by: obtaining a set of sample subgraphs classified into the given network optimization action type, and a set of ground-truth target cells in the set of sample subgraphs for the given network optimization action type, wherein the set of sample subgraphs are sampled from a training dataset for the GNN; determining, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given network optimization action type with a plurality of candidate contribution thresholds; determining respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells; and selecting, from the plurality of candidate contribution thresholds, the contributionthreshold for the given network optimization action type based on the respective matching ratios of the plurality of sets of candidate target cells.

[0195] In some embodiments, the feature information of a node in the subgraph indicates at least one of the following: key performance indicator information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node; and / or wherein the feature information of an edge between two nodes in the subgraph indicates a strength of the relationship between two cells represented by the two nodes.

[0196] In some embodiments, the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

[0197] In some example embodiments, the apparatus further comprises means for performing other operations in some example embodiments of the method 1000. In some example embodiments, the means comprises at least one processor; and at least one memory storing instructions that, when executed by the at least one processor, cause the performance of the apparatus.

[0198] In some embodiments, a computer readable storage medium having instructions stored thereon is provided. The instructions when executed by at least one processor can cause the at least one processor to carry out the functionality in accordance with any one of the embodiments described herein. In some embodiments, the computer readable medium may be a non-transitory computer readable storage medium. A computer readable storage medium may include but not limited to an electronic, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any suitable combination of the foregoing. More specific examples of the computer readable storage medium would include an electrical connection having one or more wires, a portable computer diskette, a hard disk, a read-only memory (ROM), a random access memory (RAM), an erasable programmable read-only memory (EPROM or Flash memory), an optical fiber, a portable compact disc read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0199] In some embodiments, a computer program comprising instructions is provided. The instructions, when executed by at least one processor, cause the at least one processor to carry out the functionality in accordance with any one of the embodiments described herein. In one embodiment, a carrier containing the aforementioned computer program product is provided. The carrier is one of an electronic signal, an optical signal, a radio signal, or a computer-readable storage medium (e.g., a non-transitory computer-readable medium).

[0200] Any appropriate steps, methods, features, functions, or benefits disclosed herein may be performed through one or more functional units or modules of one or more virtual apparatuses. Each virtual apparatus may comprise a number of these functional units. These functional unitsmay be implemented via processing circuitry, which may include one or more microprocessor or microcontrollers, as well as other digital hardware, which may include DSPs, special-purpose digital logic, and the like. The processing circuitry may be configured to execute program code stored in memory, which may include one or several types of memory such as ROM, RAM, cache memory, flash memory devices, optical storage devices, etc. Program code stored in memory includes program instructions for executing one or more telecommunications and / or data communications protocols as well as instructions for carrying out one or more of the techniques described herein. In some embodiments, the processing circuitry may be used to cause the respective functional unit to perform corresponding functions according one or more embodiments of the present disclosure.

[0201] While processes in the figures may show a particular order of operations performed by certain embodiments of the present disclosure, it should be understood that such order is exemplary (e.g., alternative embodiments may perform the operations in a different order, combine certain operations, overlap certain operations, etc.).

[0202] Those skilled in the art will recognize improvements and modifications to the embodiments of the present disclosure. All such improvements and modifications are considered within the scope of the concepts disclosed herein.

Claims

WHAT IS CLAIMED IS:

1. A computer-implemented method (1000) for network optimization, comprising: generating (1010) a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes; classifying (1020), based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell; determining (1030) for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and determining (1040) for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell.

2. The computer-implemented method (1000) of claim 1, wherein the nodes in the subgraph comprise: a first node with a first node type representing the serving cell, the first node type indicating that the serving cell is an issue cell, and at least one second node with a second node type representing the at least one neighbor cell, the second node type indicating a neighbor cell of the issue cell; and wherein a type of feature information of the first node is different from a type of feature information of the at least one second node.

3. The computer-implemented method (1000) of claim 1 or 2, wherein generating a subgraph for the serving cell comprises: extracting, based on the serving cell, the subgraph from a graph corresponding to the communication network, the graph comprising a set of nodes and edges connecting the set of nodes, the set of nodes representing respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

4. The computer-implemented method (1000) of any of claims 1 to 3, wherein classifyingthe subgraph into at least one of a plurality of network optimization action types comprises: performing a subgraph classification on the subgraph using a graph neural network (GNN), the GNN comprising a plurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph; obtaining an output of the GNN, the output of the GNN at least comprising a plurality of labels corresponding to the plurality of network optimization action types, each of the plurality of labels indicating a predicted probability of the subgraph being classified into a corresponding network optimization action type; and determining the at least one network optimization action type for the subgraph based on the output of the GNN.

5. The computer-implemented method (1000) of claim 4, wherein the GNN at least comprises one or more of: a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing the at least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, the intermediate feature representation being determined based on the feature information of the first node and the at least one second node.

6. The computer-implemented method (1000) of any of claims 1 to 5, wherein determining the at least one target cell for each of the at least one network optimization action type comprises: for a given network optimization action type among the at least one network optimization action type, extracting an intermediate feature representation of the feature information of the nodes and the edges in the subgraph from a GNN, wherein the feature information is provided as input to the GNN for processing, and the classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation; determining respective contribution scores of the nodes in the subgraph to the given network optimization action type based on the intermediate feature representation, a contribution score of a node indicating a contribution of a cell represented by the node to the given network optimization action type; and determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes.

7. The computer-implemented method (1000) of claim 6, wherein the GNN comprising a plurality of processing layers, and the intermediate feature representation is extracted from an output of a processing layer of the GNN.

8. The computer-implemented method (1000) of claim 6 or 7, wherein determining respective contribution scores of the nodes in the subgraph to the given network optimization action type comprises: determining, based on the intermediate feature representation, respective weights of the nodes in the subgraph with respect to the given network optimization action type; applying a transformation operation on a weighted intermediate feature representation, the weighted intermediate feature representation being obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation; and determining respective contribution scores of the nodes to the given network optimization action type based on the transformed intermediate feature representation.

9. The computer-implemented method (1000) of claim 8, wherein the transformed intermediate feature representation comprises non-normalized contribution scores of the nodes to the given network optimization action type, and wherein determining respective contribution scores of the nodes based on the transformed intermediate feature representation comprises: adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and applying a normalization operation on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.

10. The computer-implemented method (1000) of claim 9, wherein adjusting the nonnormalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell comprises: in accordance with a determination that the given network optimization action type is associated with the at least one neighbor cell, replacing a non-normalized contribution score of a first node representing the serving cell with a score of zero; andin accordance with a determination that the given network optimization action type is associated with the serving cell, replacing at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell with a score of zero.

11. The computer-implemented method (1000) of any of claims 6 to 10, wherein determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes comprises: comparing the respective contribution scores of the nodes with a contribution threshold for the given network optimization action type; selecting at least one of the nodes with a contribution score exceeding the contribution threshold; and determining at least one cell represented by the at least one selected node in the subgraph as the at least one target cell for the given network optimization action type.

12. The computer-implemented method (1000) of claim 11, further comprising: determining the contribution threshold for the given network optimization action type by: obtaining a set of sample subgraphs classified into the given network optimization action type, and a set of ground-truth target cells in the set of sample subgraphs for the given network optimization action type, wherein the set of sample subgraphs are sampled from a training dataset for the GNN; determining, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given network optimization action type with a plurality of candidate contribution thresholds; determining respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells; and selecting, from the plurality of candidate contribution thresholds, the contribution threshold for the given network optimization action type based on the respective matching ratios of the plurality of sets of candidate target cells.

13. The computer-implemented method (1000) of any of claims 1 to 12, wherein the feature information of a node in the subgraph indicates at least one of the following: key performance indicator information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node; and / or wherein the feature information of an edge between two nodes in the subgraph indicates a strength of the relationship between two cells represented by the two nodes.

14. The computer-implemented method (1000) of any of claims 1 to 13, wherein the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

15. An apparatus (1100) for network optimization, comprising: at least one processor (1110); and at least one memory (1120) coupled to the at least one processor (1110), the at least one memory (1120) comprising instructions that when executed by the at least one processor (1110) implement acts comprising: generating (1010) a subgraph for a serving cell of a communication network based on determining that the serving cell is an issue cell, wherein the subgraph comprises nodes and edges connecting the nodes, the nodes representing the serving cell and at least one neighbor cell of the serving cell, respectively, and an edge between two nodes indicating a relationship between two cells represented by the two nodes; classifying (1020), based on feature information of the nodes and the edges in the subgraph, the subgraph into at least one of a plurality of network optimization action types, wherein a network optimization action type indicating at least one network optimization action to be applied to alleviate an issue in the serving cell; determining (1030) for each of the at least one network optimization action type, based on feature information of the nodes and the edges in the subgraph, at least one target cell from the serving cell and the at least one neighbor cell; and determining (1040) for each of the at least one network optimization action type, at least one network optimization action to be applied on the at least one target cell.

16. The apparatus (1100) of claim 15, wherein the nodes in the subgraph comprise: a first node with a first node type representing the serving cell, the first node type indicating that the serving cell is an issue cell, and at least one second node with a second node type representing the at least one neighbor cell, the second node type indicating a neighbor cell of the issue cell; and wherein a type of feature information of the first node is different from a type of feature information of the at least one second node.

17. The apparatus (1100) of claim 15 or 16, wherein generating a subgraph for the serving cell comprises: extracting, based on the serving cell, the subgraph from a graph corresponding to thecommunication network, the graph comprising a set of nodes and edges connecting the set of nodes, the set of nodes representing respective cells in the communication network, and an edge between two nodes indicating a relationship between two cells represented by the two nodes in the graph.

18. The apparatus (1100) of any of claims 15 to 17, wherein classifying the subgraph into at least one of a plurality of network optimization action types comprises: performing a subgraph classification on the subgraph using a graph neural network (GNN), the GNN comprising a plurality of processing layers configured to process the feature information of the nodes and the edges in the subgraph; obtaining an output of the GNN, the output of the GNN at least comprising a plurality of labels corresponding to the plurality of network optimization action types, each of the plurality of labels indicating a predicted probability of the subgraph being classified into a corresponding network optimization action type; and determining the at least one network optimization action type for the subgraph based on the output of the GNN.

19. The apparatus (1100) of claim 18, wherein the GNN at least comprises one or more of: a first processing layer for receiving feature information of a first node representing the serving cell as input, a second processing layer for receiving the feature information of at least one second node representing the at least one neighbor cell as input, and a third processing layer for receiving feature information of the edges and an intermediate feature representation as input, the intermediate feature representation being determined based on the feature information of the first node and the at least one second node.

20. The apparatus (1100) of any of claims 15 to 19, wherein determining the at least one target cell for each of the at least one network optimization action type comprises: for a given network optimization action type among the at least one network optimization action type, extracting an intermediate feature representation of the feature information of the nodes and the edges in the subgraph from a GNN, wherein the feature information is provided as input to the GNN for processing, and the classifying of the at least one network optimization action type for the subgraph is based on the intermediate feature representation; determining respective contribution scores of the nodes in the subgraph to the given network optimization action type based on the intermediate feature representation, a contributionscore of a node indicating a contribution of a cell represented by the node to the given network optimization action type; and determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes.

21. The apparatus (1100) of claim 20, wherein the GNN comprising a plurality of processing layers, and the intermediate feature representation is extracted from an output of a processing layer of the GNN.

22. The apparatus (1100) of claim 20 or 21, wherein determining respective contribution scores of the nodes in the subgraph to the given network optimization action type comprises: determining, based on the intermediate feature representation, respective weights of the nodes in the subgraph with respect to the given network optimization action type; applying a transformation operation on a weighted intermediate feature representation, the weighted intermediate feature representation being obtained by weighting, with the determined respective weights, respective feature elements of the intermediate feature representation for the nodes in the subgraph, to obtain a transformed intermediate feature representation; and determining respective contribution scores of the nodes to the given network optimization action type based on the transformed intermediate feature representation.

23. The apparatus (1100) of claim 22, wherein the transformed intermediate feature representation comprises non-normalized contribution scores of the nodes to the given network optimization action type, and wherein determining respective contribution scores of the nodes based on the transformed intermediate feature representation comprises: adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell; and applying a normalization operation on the adjusted non-normalized contribution scores, to obtain the respective contribution scores of the nodes.

24. The apparatus (1100) of claim 23, wherein adjusting the non-normalized contribution scores of the nodes based on a determination of whether the given network optimization action type is associated with the serving cell or the at least one neighbor cell comprises: in accordance with a determination that the given network optimization action type isassociated with the at least one neighbor cell, replacing a non-normalized contribution score of a first node representing the serving cell with a score of zero; and in accordance with a determination that the given network optimization action type is associated with the serving cell, replacing at least one non-normalized contribution score of at least one second node representing the at least one neighbor cell with a score of zero.

25. The apparatus (1100) of any of claims 20 to 24, wherein determining at least one target cell for the given network optimization action type based on the respective contribution scores of the nodes comprises: comparing the respective contribution scores of the nodes with a contribution threshold for the given network optimization action type; selecting at least one of the nodes with a contribution score exceeding the contribution threshold; and determining at least one cell represented by the at least one selected node in the subgraph as the at least one target cell for the given network optimization action type.

26. The apparatus (1100) of claim 25, the acts further comprising: determining the contribution threshold for the given network optimization action type by: obtaining a set of sample subgraphs classified into the given network optimization action type, and a set of ground-truth target cells in the set of sample subgraphs for the given network optimization action type, wherein the set of sample subgraphs are sampled from a training dataset for the GNN; determining, from the set of sample subgraphs, a plurality of sets of candidate target cells for the given network optimization action type with a plurality of candidate contribution thresholds; determining respective matching ratios of the plurality of sets of candidate target cells to the set of ground-truth target cells; and selecting, from the plurality of candidate contribution thresholds, the contribution threshold for the given network optimization action type based on the respective matching ratios of the plurality of sets of candidate target cells.

27. The apparatus (1100) of any of claims 15 to 26, wherein the feature information of a node in the subgraph indicates at least one of the following: key performance indicator information of a cell represented by the node, or a set of configuration parameters of the cell represented by the node; and / or wherein the feature information of an edge between two nodes in the subgraph indicates astrength of the relationship between two cells represented by the two nodes.

28. The apparatus (1100) of any of claims 15 to 27, wherein the serving cell is determined as an issue cell based on detecting that a performance degradation on the serving cell meets a degradation threshold.

29. A computer readable medium having instructions stored thereon, the instructions when executed by at least one processor (1110) cause the at least one processor (1110) to perform the method (1000) according to any of claims 1 to 14.

30. A computer program product comprising instructions which, when executed on at least one processor (1110), cause the at least one processor (1110) to perform the method (1000) according to any of claims 1 to 14.

Images