WO2019034805A1

WO2019034805A1 - Customer-centric cognitive self-organizing networks

Info

Publication number: WO2019034805A1
Application number: PCT/FI2018/050552
Authority: WO
Inventors: Muhammad ASGHAR
Original assignee: University Of Jyväskylä
Priority date: 2017-08-14
Filing date: 2018-07-20
Publication date: 2019-02-21
Also published as: FI20175725A1

Abstract

The present invention relates to a method, a system and a computer program for managing a cognitive self-organizing network comprising heterogenous wireless networks. Real-time network data is collected, the real-time network data comprising at least operations support system data (621, 631), context data (651), social networks data (641) and customer retention management data (611). The collected real-time network data is unified to form unified real-time network information comprising at least unified key performance indicators (612), unified customer mobility information (622) and unified business intelligence information (632). The unified real-time network information is knowledge mined (613). The unified and knowledge mined real-time network information is modeled for extracting system and customer behavior models (305). An up-to-date model (305) of the self-organizing network is maintained, the up-to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time. The cognitive self-organizing network is self-configured (301), self-optimized (302), self-healed (303) and self-protected (304) on basis of the up-to- date model.

Description

Customer-centric cognitive self-organizing networks

Field

The present invention relates to a method, a system and a computer program product related to cognitive self-organizing networks (CSON). More particularly, the present invention discloses a conceptual architecture for a CSON and a method, a system and a computer program product which enable cognitive, customer centric management of a CSON.

Background

4G LTE networks are well-established both in terms of handsets available and also the infrastructure. Now the eyes of the mobile telecommunication industry are turning towards the next generation of technology referred to as the 5G. The concept of 5G network refers generally to an integrated wireless network, where a number of heterogenous wireless networks and wireless access technologies are integrated to seamlessly serve the customers.

Figure 1 illustrates an exemplary heterogenous multi-layer network where different resource layers coexists. Complexity of the arrangement is manifest. A geographic area may be covered by a variety of wide area, medium area, hot spot and indoor networks. The wide area networks may comprise GSM, CDMA and/or HSPA+ networks. The medium area networks may comprise LTE, WiMAX and/or TD-LTE networks. Hot spots may comprise LTE-A and/or TD-LTE-A networks, and further the indoor networks may comprise for example WiFi, LTE Femto and/or HSPA+ Femto networks. 5G networking technology is needed because there is a growing demand for coverage, capacity and quality of service. Self-Organizing Networks (SON) will be the default mode of operation in 5G networks. In telecom industry, the network operations, hereafter referred with the terms Operations Support System (OSS) and OSS domain, and customer service functions, hereafter referred with the term Customer Retention Management (CRM) and CRM domain, are traditionally separated from each other. This separation between CRM and OSS in telecommunication industry has been a viable strategy for most telecom operators. The customer satisfaction levels and Net Promoter Score (NPS) are hugely dependent on network performance. The persistent dropped calls, weak signals, and slow data connections are some of many reasons that cause shift in customer loyalty and can build up towards a potential desire to churn. The customer's actual experience when using their handset on the network is one of the most important touchpoints, and this is ultimately driven by the network Key Performance Indicators (KPIs)

The traditional network management worked fine when network optimization was largely a manual process and it was performed by teams of Radio Frequency (RF) engineers who regularly adjusted the network parameters to achieve the optimal coverage and performance. However, now that the 5G era brings major transformation in telecom industry, it is high time to redesign the business models and to rethink business's infrastructure. This is a great opportunity to bring the CRM and OSS systems, processes and data closer together. Therefore, a new approach is crucial for deeper OSS and CRM integration to result in increased profitability.

To run a customer-centric service, network optimization needs to be based on actionable subscriber experience insights. Raw network performance metrics are just not adequate any more, especially for complex services such as VoLTE, video streaming, and over-the-top applications like WhatsApp®, nor are they able to meet the needs of highly mobile subscribers who need seamless coverage when travelling. All kinds of user behavior, such as commuting patterns, social media habits, and the types of data services accessed at different times of day, can all impact on network optimization needs. What is more, using an integrated CRM-SON solution to optimize user experience actually starts to mirror the organizational evolution that is occurring within Mobile Network Operators (MNOs).

The move towards network functions virtualization (NFV) will make it more difficult to obtain data from probes and traces and will therefore also be a factor in accelerating the move towards a more integrated service approach. On the other hand, NFV itself provides operators with more flexibility to tailor services to suit subscribers' individual needs, and this two-way relationship will also affirm the need to break out of those historical silos.

Description of the related art

US patent application publication US 2015/0017975 discloses a method for outage detection and recovery in heterogenous networks. The system is self-healing, and it relies on location information while making decisions based on measurements performed in the underlying heterogenous network. A typical self-healing process is in principle a reactive process: the process of self-healing is initiated only after a problem is detected. The problem is then diagnosed to find the root cause of the problem, and a solution is found and deployed for removing the root cause in order to reduce likelihood of like problem to occur in future. Summary

In order to attract new customers and retain the already connected customers, an operator must have to address customer needs and concerns proactively, even before problems occur. An object is to provide a method and system so as to solve the problems of siloed OSS and CRM domains by introducing a proactive, cognitive SON (CSON). The objects of the present invention are achieved with a method according to the characterizing portion of claim 1. The objects of the present invention are further achieved with a system according to the characterizing portion of claim 9 and with a computer program product according to claim 17.

The preferred embodiments are disclosed in the dependent claims.

According to a first aspect, a computer implemented method for managing a cognitive self-organizing network comprising heterogenous wireless networks is provided. The method comprises collecting real-time network data, the real-time network data comprising at least operations support system (OSS) data, context data, social networks data and customer retention management data and unifying the collected real-time network data into unified real-time network information comprising at least unified key performance indicators, unified customer mobility information and unified business intelligence information. The method comprises knowledge mining the unified real-time network information, modeling the unified and knowledge mined real-time network information for extracting system and customer behavior models and maintaining an up- to-date model of the self-organizing network, the up-to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time. The modeling comprises recognizing at least one customer with increased risk of customer churn and decreased customer mood. The method comprises enabling recording of detailed radio measurement logs for the recognized at least one customer for at least one of a defined geographical scope and a temporal scope for performing detailed diagnoses of problems experienced by the recognized at least one customer. The method comprises self-configuring, self- optimizing and self-healing the cognitive self-organizing network on basis of the up-to-date model. The self-healing comprises automatically diagnosing performance degradation experienced by the recognized at least one customer, and automatically triggering correction actions in the cognitive self-organizing network. The self-optimization comprises proactively scheduling preventive actions in the cognitive self-organizing network on basis of the predicted behavior of the self-organizing network and the predicted behavior of the recognized at least one customer. According to a second aspect, the method further comprises self- protecting the cognitive self-organizing network on basis of the up-to-date model.

According to a third aspect, the prediction of customer behavior models comprises predicting temporal and geographical customer behavior in real-time on basis of the up-to-date model.

According to a fourth aspect, the predicted customer behavior model proactively causes self-configuring and/or self-optimizing and/or self- healing the cognitive self-organizing network to handle extra network load predicted at a specific time in a defined geographical location.

According to a fifth aspect, the knowledge mining the unified key performance indicators comprises ranking key performance indicators with respect to their impact on the operational and business objectives, filtering out key performance indicators that are below a certain threshold, and relating each key performance indicator to a network parameter that has direct influence on the key performance indicator and arranging the network parameters associated with the key performance indicators in a specific order with respect to the strength of their association. According to a sixth aspect, the knowledge mining the unified business intelligence information comprises ranking the unified business intelligence information based on its impact on the business and revenue targets, filtering the unified business intelligence information for removing the customer retention management information elements that are below a certain threshold, identifying in real time cell sites that are most likely providing coverage to the specific customer in the future, the cell cites defining a geographical scope of the customer;- acquiring a current customer profile from the network and extracting the customer behavior model on basis of the filtered, unified business intelligence information, the current customer profile, and at least one of the unified customer mobility information and the geographical scope of the customer. According to a seventh aspect, the method further comprises providing updated network parameters based on the up-to-date model, validating the updated network parameters by exploiting the updated network parameters in simulation, wherein the simulation determines the impact of the updated network parameters on the network performance, and providing the updated the network parameters to the respective network elements, if the validating phase indicates that the updated network parameters will achieve the wanted network performance.

According to an eighth aspect, the method further comprises predicting customer churn probability on basis of the customer behavior model and feeding back predicted customer churn probability towards the customer retention management system.

According to another aspect, a computer program product is provided, the computer program product having instructions which when executed by a computing device or system cause the computing device or system to perform the method of managing a cognitive self-organizing network comprising heterogenous wireless networks according to any one of the above aspects.

According to an aspect, a data-processing system for managing a cognitive self-organizing network comprising heterogenous wireless networks is provided. The system comprises at least one data collection functionality configured to collect real-time network data, the real-time network data comprising at least operations support system data, context data, social networks data and customer retention management data and a unifying functionality configured to unify the collected real-time network data into unified real-time network information comprising at least unified key performance indicators, unified customer mobility information and unified business intelligence information. The system comprises a knowledge discovery and analytics functionality configured to knowledge mine the unified real-time network information, and the knowledge discovery and analytics functionality is further configured to model the unified and knowledge mined real-time network information for extracting system and customer behavior models. The modeling comprises recognizing at least one customer with increased risk of customer churn and decreased customer mood. The system comprises a CRM system configured to enable recording of detailed radio measurement logs for performing detailed diagnoses of problems experienced by the customer. The recording is enabled for the recognized at least one customer for at least one of a defined geographical scope, and a defined temporal scope. The system comprises a SON Engine configured to maintain an up-to-date model of the self-organizing network, the up-to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time and the SON Engine is further configured to self-configure, self-optimize, self-heal and self-protect the cognitive self-organizing network on basis of the up-to-date model. The self-healing comprises automatically diagnosing performance degradation experienced by the recognized at least one customer, and automatically triggering correction actions in the cognitive self-organizing network. The self-optimization comprises proactively scheduling preventive actions in cognitive self-organizing network on basis of the predicted behavior of the self-organizing network and the recognized at least one customer. According to a further aspect, the SON Engine is further configured to self- protect the cognitive self-organizing network on basis of the up-to-date model.

According to another aspect, the knowledge discovery and analytics functionality is further configured to rank key performance indicators with respect to their impact on the operational and business objectives, to filter out key performance indicators that are below a certain threshold, and relate each key performance indicator to a network parameter that has direct influence on the key performance indicator and to arrange the network parameters associated with the key performance indicators in a specific order with respect to the strength of their association.

According to yet another aspect, the knowledge discovery and analytics functionality is further configured to rank the unified business intelligence information based on its impact on the business and revenue targets, to filter the unified business intelligence information for removing the customer retention management information elements that are below a certain threshold, to identify in real time cell sites that are most likely providing coverage to the specific customer in the future, the cell cites defining a geographical scope of the customer, to acquire a current customer profile from the network, and to extract the customer behavior model on basis of the filtered, unified business intelligence information, the current customer profile, and at least one of the unified customer mobility information and the defined geographical scope of the customer.

According to a further aspect, the system comprises a validation functionality configured to receive updated network parameters based on the up-to-date model, to validate the updated network parameters by exploiting the updated network parameters in simulation, wherein the simulation determines the impact of the updated network parameters on the network performance, and to provide the updated the network parameters to the respective network elements, if the validating phase indicates that the updated network parameters will achieve the wanted network performance.

According to a yet further aspect, the SON engine is further configured to predict customer churn probability on basis of the customer behavior model, and to feed the predicted customer churn probability back towards the customer retention management system.

According to another aspect, a computer readable medium is provided, the computer readable medium having stored thereon instructions which when executed by a computing device or system cause the computing device or system to perform :

- collecting real-time network data, the real-time network data comprising at least operations support system data, context data, social networks data and customer retention management data;

- unifying the collected real-time network data into unified real-time network information comprising at least unified key performance indicators, unified customer mobility information and unified business intelligence information;

- knowledge mining the unified real-time network information;

- modeling the unified and knowledge mined real-time network information for extracting system and customer behavior models;

- maintaining an up-to-date model of the self-organizing network, the up- to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time; and

- self-configuring, self-optimizing, self-healing and self-protecting the cognitive self-organizing network on basis of the up-to-date model.

The present invention is based on the main idea that the mobile operator should not only have a holistic view of its customers, but the mobile operator should also track its customers' experiences who are using its network and utilize this information for improving network operation and thus customer experience perceived by the customers using the network. The mobile operator needs to fully exploit the valuable customer data it can obtain from its network and services hosted in the network. In a proactive approach the mobile operator may access and act on all customer data they have in real time, whether from the network, OSS, BSS or CRM domains, and the network may be proactively managed on basis of such real time customer data. Also, the customer data may be used together with the predictive indicators based on this data.

The present invention has the advantage that an improved level of insight enables mobile operators to add value to their customers through improved customer experience, network performance and efficiency. Thus, mobile operators are capable of differentiating themselves from the new breed of over-the-top internet and device competitors that simply cannot access the same information. The 5G CSON is a modular, scalable, extensible, and smart network management architecture which explores the integration of cutting-edge technologies widely recognized as key enabling technologies for 5G systems, such as Software-Defined Networking (SDN), Network Function Virtualization (NFV), Cognitive and Cloud Computing, and Artificial Intelligence to implement an autonomic network management system addressing a number of essential management tasks, including but not limited to automated network monitoring, autonomic network maintenance, automated deployment of network tools, and automated network service provisioning.

In addition to self-configuring, which is a known feature of self-organizing networks (SON), the 5G CSON design focuses on three major network management areas:

1) Self-healing

2) Self-optimization

3) Self-protection

5G CSON introduces intelligence, self-organizing, and autonomic capacities to 5G networks, providing an open environment to foster innovation and decrease the CAPEX and OPEX of new applications. Brief description of the drawings

In the following the invention will be described in greater detail, in connection with preferred embodiments, with reference to the attached drawings, in which

Figure 1 is an illustration of an exemplary heterogenous multi-layer network.

Figure 2 illustrates a vision of a CSON solution for 5G networking.

Figure 3 illustrates a conceptual architecture of a 5G Cognitive Self- Organizing Network. Figure 4 illustrates content and categorization of CRM information.

Figure 5 illustrates an exemplary 5G CSON architecture topology.

Figure 6 illustrates how customer intelligence is established in the network in real-time.

Figure 7 illustrates use of CRM information for predicting user behavior. Figure 8 illustrates an exemplary embodiment of predicting user behavior in the CSON.

Figure 9 illustrates an exemplary node device.

Detailed description

The following embodiments are exemplary only. Although the specification may refer to "an", "one", or "some" embodiment(s), this does not necessarily mean that each such reference is to the same embodiment(s), or that the feature only applies to a single embodiment. Single features of different embodiments may be combined to provide further embodiments.

In the following, features of the invention will be described with a simple example of a system architecture in which various embodiments of the invention may be implemented. Only elements relevant for illustrating the embodiments are described in detail. Various implementations of the information system comprise elements that are suitable for the example embodiments and may not be specifically described herein. A functionality refers to one or more nodes performing specific steps of a method. Nodes are parts of a computer network, in other words part of a set of computers connected together for the purpose of sharing resources. Nodes may be physical or virtual.

Examples of a method, a system, a computer readable medium, and a computer program product for managing a cognitive self-organizing network. The traditional approach for most telecom operators has been to put up a gap between Customer Retention Management (CRM) and Operation Support System (OSS) management. With the coming of new technologies of 4G and 5G, the telecom industry already undergoes a major transformation in view of communications and content moving to IP (Internet Protocol) packet switched domain. This may be a beginning to a redesign of the business models and the time to rethink telecom businesses infrastructure. There is a great opportunity in bringing together the CRM and OSS systems. An architectural change with a modular approach is crucial for deeper OSS and CRM domain integration, which may result in increasing profitability for telecom operators.

The figure 2 shows a vision of the CSON solution for 5G networks which works as a basis for the invention. Traditional Customer Retention Management (CRM) is passive in nature and does not match the dynamic needs of 5G users. For example, most of the information from charging and billing systems is inherently static and does not tell the mobile operator much about the customer mood and experience. If a customer has experienced poor network performance, why the mobile operator should wait for the person to contact helpline and complain. In 5G networks there will be a massive amount of small cells deployed and it would be very difficult to manage such a complex and huge network with traditional SON and with passive CRM. Mobile operator should be able to identify such issues and it needs to generate proactive response.

In traditional network management, the Customer Retention Management CRM 210 focusing on operations related to customer interface, comprising for example customer profiles 211, billing and charging functionalities 212, business intelligence functionalities 213 and order management functionalities 214, is predominantly separated from the OSS domain 220 functionalities focusing on operations of the network itself. Examples of OSS domain 220 functionalities are location services 221, policy management 222, quality of service management 223 and service and usage data collection 224. With the advent of SON, however, this 'silo' approach is no longer adequate, and there needs to be much closer cooperation between departments. Service issues need to be rapidly addressed before they result in loss of customers or damage to brand reputation. This is a considerable challenge when new mobile applications are being launched almost daily, making network traffic and user mobility profiles ever more difficult to predict. In the CSON, the customer related data provided by the CRM and OSS domains is therefore combined to provide a customer centric approach both to customer management and to network management. This includes a holistic customer view with insights into customer experience and even customer mood, as well as insights into network performance and efficiency - not only as measured from the network but also as the customer experiences these.

By combining customer related information available from the CRM domain 210 with information on the OSS domain 220, the entire network management approach will become customer centric. The network management thus takes into account a holistic customer view including insights into customer experience and even customer mood and utilizes the insight of the mobile operator into network performance and efficiency to achieve a network with coverage, capacity and quality of service that best serves the customers while not wasting resources. To highlight the new approach of CSON, we prefer using the term customer typically used in the CRM domain instead of term user typically used in the OSS domain. However, either term may be used. Figure 3 illustrates a conceptual architecture of a 5G Cognitive Self- Organizing Network CSON. Main parts of the architecture are a SON Engine (SON-E) 300, a localization system (LOC) 310 and a real-time CRM (RT-CRM) 320. Within the localization system 310, the calculation of a user terminal's geographical position can be done using techniques based on received power measurements, Angle of Arrival (AoA), Time of Arrival (ToA) or Time Difference of Arrival (TDoA) can be used.

Taking into account the short distances and indoor nature of femtocell deployments, user's position in an indoor scenario is preferably estimated using received power measurements (Received Signal Strength, RSS). The localization system 310 receives Received Signal Strength (RSS) measurements from the user terminals involved in the positioning process in an indoor scenario. It evaluates these measurements and estimates the positions of the user terminals.

In 2G networks, such as GSM and CDMA, the Received Signal Level (RxLev) can be analyzed. In 3G networks, the Common Pilot Channel (CPICH) Received Signal Code Power (RSCP) can be measured. In LTE networks, Reference Signal Received Power (RSRP) is used. Therefore, Received Signal Strength (RSS) may be used as a general term for all generations of cellular mobile communications.

Although the estimation of the user terminal position based on RSS can be achieved by means of tri-lateration or signal pattern recognition, the complex propagation conditions in indoor environments reduce the accuracy of systems based on tri-lateration, and therefore techniques based on signal pattern recognition, known as fingerprint, may also be considered. To enhance accuracy in the position assessment obtained using RSS, information collected from the sensors of the user terminal can be additionally considered. Such sensors may include sensors configured to provide information on orientation and/or acceleration. Examples of such sensors are for example accelerometers, gyroscopes and magnetometers. After estimating the position of the user terminal, the localization system 310 issues a notification which contains position of the user terminal among other information such as RSS values. The notification is sent to the SON engine 300 as a well-defined Javascript Object Notation (JSON) message. The localization system 310 may comprise Minimization of Drive Test (MDT) functionality 311. The 3GPP standardized MDT concept allows the generation of user terminal traces, i.e., sets of samples obtained from a particular mobile terminal, which can include where and when such terminal related samples were obtained. These traces are commonly analyzed a posteriori by human experts supported by visual graphic tools. The use of such traces for automatic healing purposes may be configured to be a part of the localization system 310.

The localization system 310 may further construct and/or maintain a radio frequency (RF) fingerprinting database 312. RF fingerprinting is a process for the proper identification of wireless devices in mobile and wireless networks. It also makes use of the MDT functionality to collect location info and other related info in real-time.

In the presented CSON solution, we envision that when a customer faces any problem due to poor network performance (for example poor coverage etc.) and calls to customer service for making a complaint, the customer service representative already has a holistic view of the customer in real-time both the customer history of using services as well as network performance through the real-time Customer Retention Management CRM 320 part of the CSON architecture. The real-time CRM part 320 thus beneficially includes both customer profile 321 and customer experience information 322, which may be stored for example in one or more databases. The CMR part maintains a holistic view of the customer in real-time. This may include for example customer complaints history, customer problems area, location and time of problems experienced by the customer as well as face by the customer, to name a few examples. The customer problems area refers to a geographical area in which the customer is facing problems. This may be for example an area where there is a hole in the network coverage or the coverage is weak so that customers often have their calls dropped in this area.

The CSON solution helps to improve the Customer Retention Management CRM for customer satisfaction at one hand and improves network efficiency on the other hand by speeding up the network self- configuration, network self-optimization and network self-healing processes.

Context information preferably includes position, orientation and other information. Context information may comprise information and the influence of nearby devices which can be used in device-to-device communication. Context information may be collected from the user terminals, which may provide for example information on position, call logs, network connections. Context information may also be collected from the cellular network, in which case the context information may comprise network status information and network configuration information. Context information may also be collected from the surrounding environment. In such case, context information may comprise for example weather reports, information on changes in surrounding infrastructure such as new constructions and/or buildings, temporary changes in the environment such as events in the city or in an indoor facility, information on arriving or leaving trains, ferries or airplanes and so on. Context information of the user can be used in the SON engine 300 and in the real-time CRM 320. For example, the SON engine 300 may utilize any received or obtained context information for self-configuration, self-optimization and self-healing. For example, the real-time CRM part 320 may utilize current user geographical location and the current quality of service (QoS) received by the user as context information. The QoS may be represented by network information, such as Reference Signal Received Power (RSRP), Reference Signal Received Quality (RSRQ) and/or Channel Quality Indicator CQI. The QoS may alternatively or in addition be considered in broader perspective. In such case, the QoS may be represented for example in terms of call setup rate, call drop rate etc. The SON Engine 300 may be configured to perform any selection of tasks of self-configuration (301), self-optimization (302), self-healing (303) and self-protection (304). A model (305) comprising information on both the heterogenous network and customers (clients) is generated and continuously updated to enable performing these tasks. For this, data is modeled to extract system and user behavior models on which the SON Engine 300 can perform SON functions. Traditional static methods for modeling the network and users is not useful because most of the SON functions deal with dynamic network and user behavior. Therefore, in the CSON, the latest artificial intelligence methods such as deep learning are employed to correctly model the dynamic network and customer behavior.

The term self-configuration (301) means for example bringing new base stations to network with minimum human intervention. There are three main steps for self-configuration, which may be referred to as auto- connectivity, auto-commissioning and dynamic radio configuration. 1) Auto-connectivity refers to the stage after physical cabling has been coupled and power has been switched on. It includes automatic connection to Operations, Administering and Management (OAM) system with IP address allocation and authentication for secure connectivity. Auto-connectivity may also comprise site identification to select correct Software and configuration settings. 2) Auto-commissioning refers to that when a Network Equipment (NE) leaves from factory it has only minimum software (SW) and parameters. Once auto-commissioning is made correct SW build and a Configuration Management (CM) database is downloaded at the NE. 3) Dynamic Radio Configuration includes features enabling omitting of detailed radio planning, reducing labor intensive manual planning and more accurate parameter setting based on measurements from actual network. Parameters are selected based on network's current configuration with neighbors (e.g. Physical Cell ID) and measurements (e.g. Automatic Neighbor Relations).

The term self-optimization (302) means utilizing the network measurements and performance indicators collected from user equipment and base stations for auto-tuning the network settings. Self-optimization may be initiated and/or performed by the SON Engine, facilitated by the up-to-date model. Self-optimization may comprise for example tuning network parameters for improving network performance experienced by one or more users.

The term self-healing (303) means automatically detecting and diagnosing performance degradation, and automatically trying to compensate cell degradations and trigger correction actions. The correction actions may be initiated and/or performed by the SON Engine on basis of the up-to-date model. Self-healing may be needed for example, if the SON network identifies a broken base station, which may be automatically fixed or compensated by actions of the SON itself. The term self-protection (404) means for example monitoring the network continuously and make an auto-connection to the OAM system and ensure secure authentication of users and service. In addition, self-protection refers to detecting and blocking any intrusion attempts from outside the network towards the sensitive network data.

The SON Engine 300 may utilize context information so that it can both react on detected changes in the context, but also to predict forthcoming context of a user terminal. Such predicted context may be utilized for configuring and optimizing the self-organizing network in advance, so that the network provides best possible user experience and quality of service. This kind of proactive CSON network management based on predicted context may be especially beneficial when the predicted context indicates that an exceptionally large network capacity will be needed in a specific location: the network may be optimized for the extra network load in advance and thus likelihood of network failures due to overload may be significantly reduced.

The CSON is customer centric, meaning that it collects, maintains and provides holistic view of customer behavior in terms of e.g., customer mobility, usage of mobile data in long term, user experience and customer mood in real-time.

Figure 4 illustrates exemplary content and categorization of CRM information. CRM information may be divided to main categories of cost, product quality and customer service data. Cost information represents costs associated with entering and maintaining a relationship with a service provider. Such cost information may comprise for example information on service prices and switching cost, if a customer would switch between different service providers (mobile operators). Product quality information represents quality of the physical product, the associated core services and the quality of the contract fit between the expectations of the customer and the provided service. For example, the product quality information may comprise information on phone service quality and phone plan quality. Customer experience represents interfaces between the customers and the telecom operator in a larger context. Customer experience information reflects ability to deal with everyday service, exceptions, public responses and the associated service provider image that is generated. Customer experience information may comprise complaint management, customer service quality information, brand image information and social norm information.

Figure 5 illustrates an exemplary 5G CSON architecture topology, which is in line with the Network Functions Virtualization (NFV) and Software Defined Network (SDN) concepts on the ETSI architecture that is determined by ETSI. The architecture topology consists of six layers: 1) Infrastructure Layer (501), 2) Virtualized Network Layer (502), 3) Control Layer (503), 4) SON Layer (504), 5) NFV Orchestration and Management functionality (505) 6) Operator's Interface Layer (506).

The Infrastructure Layer (501) represents the physical resources and the resources required for the instantiation of virtual functions (e.g. Virtual Compute, Network, and Storage) and supports the mechanisms for that instantiation. The infrastructure layer (501) may be divided to a physical sublayer (511) and a virtualized sublayer (521).

The Virtualized Network Layer (502) represents the instantiation of the virtual networking infrastructures. The Virtualized Network Layer (502) is composed of a number of virtual network functions (VNFs) interconnected in a designed topology in order to provide the functionalities required by the operator. The Control Layer (503) acquires measurements from sensors deployed through the cellular network (513) and provide new configurations and actions into the network as part of the enabling mechanisms to provide network intelligence in next generation SON networks. The SON Layer (504) provides the mechanisms to provide the network intelligence. It collects measurements from the network and other sources, uses the collected information to establish the current network behavior, detects coverage and/or capacity problems, diagnoses the network condition and decides what must be done to accomplish the system goals. The SON Engine (514) deals with the network problems using Artificial Intelligence (AI) and machine learning algorithms that will decide problem resolution actions to be taken in the network. The SON manager (534) provides recommendations for the problem resolution actions to be taken in the network. These actions may involve the services allocated on NFVs onboarding (524).

The actions to be taken defined by the SON Manager (534) are coordinated by the NFV orchestration and management functionality (505) and the corresponding virtualized infrastructure manager module (515). The NFV orchestration and management functionality (505) ensures that SDN/NFV apps are allocated with sufficient resources to perform their tasks and to execute their corrective actions. The virtualized infrastructure manager module (515) applies the actions in the virtualized sublayer (521) network elements of the infrastructure layer (501) and the virtualized network layer (502). The NFV orchestration functionality (525) enables the Orchestrator (514) with capabilities that allow the management and configuration of SDN/NFV applications. Thus, the NFC orchestrator and management functionality (505) ensures that SDN/NRV applications are allocated with sufficient resources to perform their tasks. The Operator's Interface Layer (506) (or in short Interface layer) encompasses the interface to the mobile operator. The mobile operator can monitor and control the CSON system through this interface. A general Application Programming Interface (API) (516) facilitates the interface between the Interface layer (506) and the functionalities of the CSON. Although the CSON works in a close-loop automation of optimization and healing mechanisms, the mobile operator will have the control over the final decisions and possible corrective actions on the live network. The policy and business targets are communicated to the CSON system through this interface. The Operator's Interface Layer (506) may provide a graphical dashboard (526) to the mobile operator's personnel.

Figure 6 illustrates an example on establishing customer intelligence in the network in real-time in the CSON system. In order to create a holistic picture of the network status, artificial intelligence techniques may be employed for example. In this exemplary case, we will use deep learning to intelligently establish the network status in real-time.

There is huge amount of diverse data available in the network that can be exploited. First phase is the collection of available data (601) from the network as well as context information from inside as well as outside the network. This huge amount of diverse data constitutes so called big data. Exemplary types of big data are for example CRM data (611), cell site data (621), core network data (631), social networks data (641) and context information (651).

The CRM data (611) may include the customer information, control and context information. This information can be used to perform customer centric operations for supporting the key business processes such as customer experience and retention enhancement. The CRM data (611) may comprise for example the call data records (CDR) and extended data records, accessibility Key Performance Indicators (KPIs), retainability KPIs and integrity KPIs. The accessibility KPIs may comprise for example call setup time and success rate, access failure and handover failure rate. Retainability KPIs may comprise for example call success rate, handover success, call drop ratio, IP throughput, session drop rate etc. The integrity KPIs may comprise for example speech quality, packet jitter, delay, data streaming quality, and throughput delay.

The cell site data (621) includes radio frequency (RF) measurements reported from the base station such as physical resource blocks (PRB) usage, received access requests on the Random Access Channel (RACH), number of active users per cell and preamble per cell. The cell site data (621) may also comprise results of measurements performed by the user equipment including, but not limited to Channel Quality Indicators (CQI), minimization of drive tests (MDT) measurements which contain reference signal received power (RSRP) and reference signal received quality (RSRQ) values of the serving and neighboring cells. These measurements can be exploited by CSON functions for the autonomous coverage optimization and self-healing operations.

The core network data (631) may include fault data such as historical alarms logs etc., configuration data such as device configuration records, equipment logs etc., network performance data such as link utilization, call drop rate, throughput etc., security data such as authentication, authorization and auditing.

As understood by a skilled person in the art, the above presented network data collection may occur by various network elements and devices, which may comprise physical or virtual nodes residing in, coupled to or providing services of the respective network. These network data collection means are preferably provided with one or more interfaces towards data storage means configured for storing big data. The data storage means may comprise one or more memories and/or data banks.

An up to date picture of the entire network can be gathered when diverse network data from all network sources can be combined to form mobile big data (660). This big data (660) comprises at least CRM data (611), cell site data (621) and core network data (631). The big data (660) may also comprise social network origin data (641) and context data (651).

After having an aggregated data set of big data (660), the next step is to unify the data. The unifying phase 602 may include unifying multiple received key performance indicators received from different sources and possibly in different format into more significant key performance indicators (KPIs) (612), unifying at least part of context data (651) into customer mobility information (622) and unifying at least part of the CRM data (611) into business intelligence information (632). The customer mobility information (622) preferably comprises at least current geographical location of the customer, and it may further comprise at least one predicted geographical location of the customer. In addition to CRM data (611) various other data may be utilized for generating the business intelligence information (632). For example, quality of service data may be utilized for generation of the business intelligence information. The unifying process may also be called diffusion, as the various separate pieces of data are diffused into a more significant form of information. For example, in a heterogenous 5G network different access networks may provide KPIs that are not directly uniform. The unifying process may thus unify KPI data received from different networks into directly comparable measures.

In the Knowledge Discovery and Analytics phase (603) the unified information is further processed. Knowledge mining (613) transforms the big data into meaningful unified real-time information. For example, knowledge mining (613) of the unified KPIs may include following steps:

1) ranking unified KPIs with respect to their impact on the operational and business objectives

2) filtering out unified KPIs that are below a certain threshold and associating each KPI to a network parameter that has direct influence on the KPI. Preferably, only those KPI's are associated which are not filtered out.

3) arranging the network parameters associated with the unified KPIs in a specific order with respect to the strength of their association.

The threshold referred to in step 2 relates to the KPIs impact on the business objective and revenue targets. Threshold will typically be defined by the operator, since it depends on the use case. For example, when the operational and business objective is "Coverage Improvement", the KPIs related to the coverage will be ranked higher than other KPIs. In this case, RSRP measurements are more important and ranked high. This way information that has less impact on the revenue may be filtered out. For example, those customers who are actively using different services and generate more revenue for the service provider/operators may be wanted to be excluded or included from the further analysis.

Likewise, knowledge mining may comprise mining the mobility information (622) and/or the business intelligence information (632). Modeling of the unified real-time information in phase 623 is performed to extract system and customer behavior models on which a SON engine can perform SON functions using the key unified KPIs (612), the mobility information (622) and the business intelligence information (632) either directly or after knowledge mining. The traditional static methods are not useful because most of the CSON functions deal with dynamic network and customer behavior. Therefore, in CSON the latest artificial intelligence methods such as deep learning must be employed to correctly model the dynamic network and customer behavior.

Concept-drifting in phase (633) is performed so that the legitimate network changes may affect the accuracy of the network behavior model (305). Therefore, the concept-drift functionality will be triggered to tune the model (305) with the latest legitimate changes when such changes have occurred in the unified real-time information. This includes detecting need for concept-drift and retraining of the model (305).

The SON Engine (300) maintains the up to date model (305) for the prediction of the network behavior and customer behavior in real-time on basis of the modeled real-time information. The SON engine 300 runs different SON functions to produce new network parameters that are optimal for the network. The SON engine (300) comprises the self- configuration (301), self-optimization (302), self-healing (303) and self- protection (304) functionalities here illustrated with a common box of self- X functions. The SON engine also comprises policy (306) and coordination functions (307). The policy function (306) includes a set of objectives that operators can define with different priorities for the KPIs. For example, allocating the highest priority to maintain the KPI of the ratio of dropped calls due to excessive load, for example, below 2%, while trying to keep another KPI below for example 3% with lower priority. The coordination function (307) is configured to perform coordination between different SON functionalities.

The functions of the SON engine (300) supports the high-level business objectives, which include but are not limited to providing network coverage, capacity and quality of service that meets the needs of the customers effectively.

The validation functionality (606) feeds back the changes proposed by the SON Engine (300) in the form of new network parameter values which may be deployed in the network after validation. The validation may be performed by determining the impact of the new parameter value(s) on the network performance by simulation. The simulated network behavior may be exploited to determine the impact of new network parameter. The validation functionality may be implemented as a part the NFV orchestration and management functionality (505).

Figure 7 illustrates an exemplary process for implementing use of CRM information for predicting user behavior.

In the first phase 701, CRM data is acquired. The CRM data may be ranked in the phase 702 based on its impact on the business and revenue targets. A filtering process may be applied in the phase 703 to remove the elements that are below a certain threshold. Cell sites are identified in real-time in the phase 704 that are most likely providing coverage to the specific customer, thus defining a geographical scope of the customer and also reflecting the current and the most likely predicted geographical position of the customer. The customer geographical location may be acquired from the customer mobility information, which may be at least partially based on context information. The customer's current profile is also acquired from the network in the phase 705. As understood in the art, acquiring the customer profile in phase 705 is independent from handling of CRM data in that sense that it may be acquired also before or during the CRM processing phases 701-703 and/or during or before the cell site identification phase 704. In phase 706, a model of the customer behavior is generated. In phase 707, the prediction model is used that takes at least the geographical scope (including coordinates of cell sites most likely serving the customer), temporal scope (quality of service in real-time) of the customer, historical CRM data and the customer profile as input for predicting customer behavior. The customer behavior prediction module may perform prediction of customer behavior in realtime. In addition, as illustrated with phase 708, the model may provide the most likely failures occurring in the customer service in the future. In phase 709, the possible causes of the future failures are analyzed, thus enabling proactive prevention of such failures by creating a forecast for future problems in phase 710 and using the self-configuration, self- optimizing and/or self-protection functions in phase 711 for scheduling preventive actions.

Figure 8 illustrates an exemplary embodiment of predicting user behavior in the CSON. The in phase 701, CRM data is acquired. The acquired CRM data may comprise for example customer profile, customer experience information, churn probability, past complaints and a customer mood indicator. The CRM data may comprise both current and historical data. The CRM data is ranked in phase 702 based on its impact on the business impact and revenue targets and filtered in phase 703 to remove the elements from the ranked CRM data that are below a certain threshold. The cell sites are identified in phase 704 that are most likely providing coverage to the specific customer at real-time. The customer is typically served by a single cell at a time. However, the customer typically receives signals from other, neighboring cells, which may belong to any layer of the heterogenous multi-layer network. The most likely cells, which term refers to those cells which will most likely be providing coverage to the customer in the future, are identified among the neighboring cells. The customer (user) current profile is also acquired from the network in phase 705. The prediction model is generated by the SON Engine in phase 706 on basis of the acquired data preferably takes into account the geographical scope (including coordinates of relevant cell sites), temporal scope (including Quality of Service in real-time) of the customer, historical CRM data and customer profile as input. The customer behavior prediction module 707 provides for example a customer churn probability prediction (707a) in real-time. Based on the churn probability prediction, feedback may be provided towards the CRM system in phase 712, which enables the operator to take a proactive approach for preventing customer churn. For example, the customer churn probability prediction may show in the CRM system in form of a customer mood indicator (713). In addition to predicting customer behavior, the prediction module may comprise a failure prediction function (707b) that provides the most likely failures in the future (708) and a diagnosis function (707c) which diagnoses the failures' possible causes (709). Based on the predictions, future problem may be forecasted in phase 710 and preventive actions may be scheduled proactively in phase 711.

The proactive customer centric approach may be further enhanced by enabling recording of detailed radio measurement logs for a customer who's churn probability (712) is high or who's customer mood indicator (713) shows an increased likelihood for customer churn. Results of the more detailed radio measurement logs may be included in the up-to-date model. Such more detailed radio measurement logs may be enabled for a specific customer for a defined geographical scope, in other words, in a particular coverage area. This is especially beneficial if the more general customer and network model indicate a problem area for the specific customer. Further, other users in the same area may be selected for such more detailed radio measurements, and the results of measurement performed for such other users may be included in the detailed radio measurement logs. If majority of customers face problems in specific location, for example bad network coverage, call drops because of overloaded network and so on, then the system will trigger SON self- optimization or self-healing actions to adapt the network to satisfy the customers in real time. SON self-optimization actions could be for example increasing transmission power of base stations or changing mobility load balancing parameters. Likewise, more detailed radio measurement logs may be enabled for a defined temporal scope. This means for example that the radio measurement logs may be enabled for the particular selected user for a predetermined time period. The recording of detailed radio measurement logs may also be enabled for a combination of a defined geographical scope and a temporal scope. Further, other users in the same area during the same time period may be selected for such more detailed radio measurements, and the results of measurement performed for such other users may be included in the detailed radio measurement logs. Such more detailed radio measurements for other users may also be enabled for a defined temporal scope, which may be the same temporal scope as for the recognized specific user with increased risk of customer churn or decreased customer mood. This way the analysis of the problems may be improved, and the right corrective actions may be facilitated. Such detailed radio measurement logs may enable more detailed diagnoses of problems such as weak coverage, call drops or handover drops experienced by the customer. The detailed radio measurement log recording may be enabled for example in the CRM system in response to detecting increased or more than average risk of customer churn, or in response to detecting decreased or less than average customer mood. The detailed radio measurement log recording may be enabled by the mobile operator's customer care, or it may be automatically triggered on basis of the risk of customer churn reaching a preset limit value or the customer mood falling below a preset limit value. The CRM system may provide information that can be utilized for recognizing a specific customer for whom such detailed radio measurement logs may be needed. Such information from the CRM system may comprise for example customer ID, his/her past complaints, real-time complaints, problems faced by the customer, history of customer churn and/or customer profile including mobility information. Information collected for the specific, recognized customer with the detailed radio measurement logs may comprise for example mobile terminal measurements, location information, context information, information about the customer's surroundings, including other users nearby, traffic load nearby. This information may then be used by the SON engine as part of the up-to-date model to make a diagnosis on the possible problems. The SON engine preferably provides feedback about tracked customers back to the CRM system to improve CRM system capability to manage customers, update the customer satisfaction and customer churn probability.

The Inference engine functionality (810) performs detection and diagnosis of problems by comparing current KPI distributions with the contextualized profiles obtained from the context aggregation functionality. As a result of a diagnosis performed in the inference engine (810), customer mood indicator may be provided. The record update functionality (811) stores historical KPI measurements and keeps updated measurements. The network association functionality (812) performs the task of identifying of cell sites that are most likely serve the customers.

Figure 9 discloses an exemplary node device (90). As the node refers to a physical or virtual computational entity capable of managing virtual nodes associated with it. The computational entity may be a device capable of handling data. It may be a server device, computer or like running a chat application or a game application etc. The node device (90), comprises a memory (91) for storing information relating e.g. to the virtual nodes associated with it, instructions how to handle messages etc. The memory (91) may comprise volatile or non-volatile memory, for example EEPROM, ROM, PROM, RAM, DRAM, SRAM, firmware, programmable logic, etc.

The node device (90) further comprises one or more processor units (92) for processing the instructions and running computer programs and an interface unit (93) for sending and receiving messages.

It is apparent to a person skilled in the art that as technology advanced, the basic idea of the invention can be implemented in various ways. The invention and its embodiments are therefore not restricted to the above examples, but they may vary within the scope of the claims.

Glossary of abbreviations

AI Artificial Intelligence

API Application Programming Interface

BSS Business Support System

CM Configuration Management

CDR Call Data Record

CQI Channel Quality Indicator

CSON Cognitive Self-Organizing Network

CRM Customer Retention Management

JSON Javascript Object Notation

KPI Key Performance Indicator

MNO Mobile Network Operator

NE Network Equipment

NF Network Function

NFV Network Functions Virtualization

NPS Net Promoter Score

OAM Operations, Administering and Management

OSS Operations Support System QoS Quality of Service

RACH Random Access Channel

RF Radio Frequency

RSRP Reference Signal Received Power RSRQ Reference Signal Received Quality RSS Received Signal Strength

SON Self-Organizing Network

SW Software

Claims

1. A computer implemented method for managing a cognitive self- organizing network comprising heterogenous wireless networks, characterized in that the method comprises:

- collecting real-time network data, the real-time network data comprising at least operations support system (OSS) data, context data, social networks data and customer retention management (CRM) data;

- unifying the collected real-time network data into unified realtime network information comprising at least unified key performance indicators, unified customer mobility information and unified business intelligence information;

- knowledge mining the unified real-time network information;

- modeling the unified and knowledge mined real-time network information for extracting system and customer behavior models, wherein the modeling comprises recognizing at least one customer with increased risk of customer churn and decreased customer mood;

- enabling recording of detailed radio measurement logs for the recognized at least one customer for at least one of a defined geographical scope and a temporal scope for performing detailed diagnoses of problems experienced by the recognized at least one customer;

- maintaining an up-to-date model of the self-organizing network, the up-to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time; and

- self-configuring, self-optimizing and self-healing the cognitive self-organizing network on basis of the up-to-date model, wherein the self-healing comprises automatically diagnosing performance degradation experienced by the recognized at least one customer, and automatically triggering correction actions in the cognitive self- organizing network, and wherein the self-optimization comprises proactively scheduling preventive actions in the cognitive self- organizing network on basis of the predicted behavior of the self- organizing network and the predicted behavior of the recognized at least one customer.

2. The method of claim 1, wherein the method further comprises self- protecting the cognitive self-organizing network on basis of the up-to- date model.

3. The method of any of claims 1 to 2, wherein the prediction of customer behavior models comprises:

- predicting temporal and geographical customer behavior in realtime on basis of the up-to-date model.

4. The method of any of claims 1 to 3, wherein the predicted customer behavior model proactively causes self-configuring and/or self- optimizing and/or self-healing the cognitive self-organizing network to handle extra network load predicted at a specific time in a defined geographical location.

5. The method of any of claims 1 to 4, wherein the knowledge mining the unified key performance indicators comprises:

- ranking key performance indicators with respect to their impact on the operational and business objectives;

- filtering out key performance indicators that are below a certain threshold, and relating each key performance indicator to a network parameter that has direct influence on the key performance indicator; and - arranging the network parameters associated with the key performance indicators in a specific order with respect to the strength of their association.

6. The method of any of claims 1 to 5 wherein the knowledge mining the unified business intelligence information comprises:

- ranking the unified business intelligence information based on its impact on the business and revenue targets;

- filtering the unified business intelligence information for removing the customer retention management information elements that are below a certain threshold;

- identifying in real time cell sites that are most likely providing coverage to the specific customer in the future, the cell cites defining a geographical scope of the customer;

- acquiring a current customer profile from the network; and

- extracting the customer behavior model on basis of the filtered, unified business intelligence information, the current customer profile, and at least one of the unified customer mobility information and the current geographical scope of the customer.

7. The method of any of claims 1 to 6, further comprising :

- providing updated network parameters based on the up-to-date model;

- validating the updated network parameters by exploiting the updated network parameters in simulation, wherein the simulation determines the impact of the updated network parameters on the network performance; and

- providing the updated network parameters to the respective network elements, if the validating phase indicates that the updated network parameters will achieve the wanted network performance.

8. The method of any of claims 1 to 7, further comprising :

- predicting customer churn probability on basis of the customer behavior model; and

- feeding back predicted customer churn probability towards the customer retention management system.

9. A data-processing system for managing a cognitive self-organizing network comprising heterogenous wireless networks, characterized in that the system comprises:

- at least one data collection functionality configured to collect realtime network data, the real-time network data comprising at least operations support system (OSS) data, context data, social networks data and customer retention management (CRM) data;

- a unifying functionality configured to unify the collected real-time network data into unified real-time network information comprising at least unified key performance indicators, unified customer mobility information and unified business intelligence information;

- a knowledge discovery and analytics functionality configured to knowledge mine the unified real-time network information, wherein the knowledge discovery and analytics functionality is further configured to model the unified and knowledge mined realtime network information for extracting system and customer behavior models, wherein the modeling comprises recognizing at least one customer with increased risk of customer churn and decreased customer mood; - a CRM system configured to enable recording of detailed radio measurement logs for performing detailed diagnoses of problems experienced by the customer, wherein the recording is enabled for the recognized at least one customer for at least one of a defined geographical scope, and a defined temporal scope;

and

- a SON Engine configured to maintain an up-to-date model of the self-organizing network, the up-to-date model being configured to predict behavior of the self-organizing network and customer behavior in real-time, wherein the SON Engine is further configured to self-configure, self-optimize and self-heal the cognitive self-organizing network on basis of the up-to-date model, wherein the self-healing comprises automatically diagnosing performance degradation experienced by the recognized at least one customer, and automatically triggering correction actions in the cognitive self-organizing network, and wherein the self-optimization comprises proactively scheduling preventive actions in cognitive self-organizing network on basis of the predicted behavior of the self-organizing network and the recognized at least one customer.

10. The system according to claim 9, wherein the SON Engine is further configured to self-protect the cognitive self-organizing network on basis of the up-to-date model.

11. The system according to any of claims 9 to 10, wherein the prediction of customer behavior models comprises:

12. The system according to any of claims 9 to 11, wherein the predicted customer behavior model proactively causes self-configuring and/or self- optimizing and/or self-healing the cognitive self-organizing network to handle extra network load predicted at a specific time in a defined geographical location.

13. The system of any of claims 9 to 12, wherein the knowledge discovery and analytics functionality is further configured to:

- rank key performance indicators with respect to their impact on the operational and business objectives;

- filter out key performance indicators that are below a certain threshold, and relate each key performance indicator to a network parameter that has direct influence on the key performance indicator; and

- arrange the network parameters associated with the key performance indicators in a specific order with respect to the strength of their association.

14. The system of any of claims 9 to 13 wherein knowledge discovery and analytics functionality is further configured to:

- rank the unified business intelligence information based on its impact on the business and revenue targets;

- filter the unified business intelligence information for removing the customer retention management information elements that are below a certain threshold;

- identify in real time cell sites that are most likely providing coverage to the specific customer in the future, the cell cites defining a geographical scope of the customer;

- acquire a current customer profile from the network; and - extract the customer behavior model on basis of the filtered, unified business intelligence information, the current customer profile, and at least one of the unified customer mobility information and the defined geographical scope of the customer.

15. The system of any of claims 9 to 14, further comprising a validation functionality configured to:

- receive updated network parameters based on the up-to-date model;

- validate the updated network parameters by exploiting the updated network parameters in simulation, wherein the simulation determines the impact of the updated network parameters on the network performance; and

- provide the updated network parameters to the respective network elements, if the validating phase indicates that the updated network parameters will achieve the wanted network performance.

16. The system of any of claims 9 to 15, wherein the SON engine is further configured to:

- predict customer churn probability on basis of the customer behavior model; and

- feed predicted customer churn probability back towards the customer retention management system.

17. A computer program product having instructions which when executed by a computing device or system cause the computing device or system to perform the method of managing a cognitive self-organizing network comprising heterogenous wireless networks according to any one of claims 1 to 8.