The goal of ICT is to provide efficient and reliable smart products and services, completely free for customers, to create greater value
From mechanization and electrification to automation and intelligence, tools are doing more and more things, and people are less and less involved. As technology develops, human beings are increasingly focusing on creative work, and delegating repetitive work to tools and machines. After the three industrial revolutions of mechanization, electrification, and automation, we are now facing the challenge of intelligence.
Consider transportation as an example. In the era of mechanization, people invented bicycles, increasing our travel speed from about 5 kmph to 20 kmph.
In the age of electrical engineering, with the emergence of steam locomotives, diesel locomotives, and electric locomotives, travel speeds increased to about 100 kmph. Constant oil and power supply enable endless running of vehicles.
In the automation era, automatic cars make driving simpler and navigation systems have become more advanced.
In the upcoming smart era, self-driving will enhance collaboration between people and vehicles, and some or most manual operations will be automated. Driving efficiency will be improved and traffic accidents and fatalities will be greatly reduced. Moreover, new energy and smart manufacturing will significantly reduce environmental pollution and provide a better and more harmonious living environment.
The ICT industry is also facing challenges posed by the fourth, "smart" industrial revolution:
- How do we use 5G, cloud, and IoT to build an intelligent big data platform?
- How do we use smart chips and algorithms to provide intelligent solutions for various industries?
- How do we stay close to customers and provide them with smart applications that are readily accessible, secure, and reliable?
To provide smart services in ICT, a smart service platform with mutual, self-service and automation capabilities must be offered based on smart customer service
The main objective of mutual services is to solve the problem of "understanding what's been said." A QABot with question and answer management capabilities is provided to replace agents, solve most consultation problems, and transfer complex consultation problems and fault problems to manual service personnel.
The main objective of self-service is to provide KGBots with "knowledge cloud + knowledge graphing" capabilities to replace domain experts, in order to solve complex consultation problems and simple fault problems, and transfer complex faults to agents.
The main objective of automation is to provide TaskBots with "device monitoring + intelligent O&M" capabilities to replace operation scripts and solve complex fault problems in a semi-automated manner.
The mutual QABot can respond to customers 24/7 and solve simple consultation problems, but it is not so smart
The Information Retrieval-based (IR-based) QABot can segment user queries and sort questions by weight such as TF-IDF to obtain 200 or more candidate results. It then performs deep learning multi-dimensional rearrangement on the candidate results according to factors including literal meaning, popularity, and semantic meaning. If the score of the first search result is high, the QABot will forward the result to the customer. Otherwise, it transfers the problem to the agent.
The Q&A management platform manages Q&A categories and knowledge points, analyzes big data regarding user questions, and adds answers to questions without search results.
The QABot utilizes a knowledge base and Q&A matching engine to respond to customer questions 24/7. For repeated questions or existing questions in the knowledge base, the answers are very accurate.
In actual ICT services, iKnow and customer service personnel (engineers) work together to serve customers. Robots are mainly used to solve simple and repetitive consulting problems, thereby ensuring answer accuracy and improving customer satisfaction. For complex consultation problems or faults, our customer or partner can contact customer service personnel and Huawei engineers. Customer service personnel and Huawei engineers can access the customer network and make decisions to solve the problems if necessary.
However, there are two issues:
Humans can easily understand and deduce a complex question. For example, question A = question B + question C. This is very easy for humans, but it is very difficult for the QABot to deduce and answer these questions.
If question A is described differently by the customer and the expert, and the QABot does not provide the bridge for the customer and the expert to communicate, the problem may still be difficult to solve for the customer even after expert explanation.
Without effective knowledge reasoning and generalization capabilities, the QABot always offers awkward answers and does not act intelligently.
The GBot is more intelligent, as it supports problem reasoning and knowledge accumulation, and solves complex consultation problems
The knowledge graph (KG) is a structured semantic knowledge base, and is used to describe concepts and relationships in a physical world with symbols. A basic unit of the knowledge graph is an entity - relationship - entity triplet, and the entity - entity attributes - value pair triplet. Entities are interconnected through relationships, forming a knowledge mesh. The knowledge graph enables the web to change from web page links to concept links. Users can search for information by topics instead of character strings. Based on the knowledge graph, the search engine can provide structured knowledge to users in graphics. Therefore, users do not need to browse a large number of web pages, and can accurately locate and obtain knowledge. For example, is Cristiano Ronaldo one of the best football players? C. Ronaldo won the Golden Ball, which is one of the most influential football awards, so we can infer that he is one of the best football players.
(Source: Knowledge Graph and Cognitive Intelligence --- Xiao Yanghua)
The KGBot can use structured knowledge to perform knowledge reasoning. It provides multiple rounds of questions and answers to clarify user context and solve complex consultation problems for customers.
The ICT knowledge graph platform includes the product, MiB, password, alarm nodes and relationships. Compared with traditional QA search, KGBot supports the following functions:
For standard questions, standard answers can be offered without inference. For example, when a user queries for the password (what is the password for S5720 SI?), the answer can be provided immediately.
For non-standard questions for which no standard answers can be matched, reasoning and inference are required. In the same password query scenario, if a user asks what the password for S5700 is, KGBot can use the PBI and password tool to conduct reasoning, and provides final answers through multiple rounds of questions and answers based on numerous knowledge results. For example, S5700 series products contain S5720 and S5720 SI. KGBot can determine what specific product the customer intends to ask about through multiple rounds and questions.
It can be likened to entering a shop and asking for a cup of cola, the waiter may ask whether you want Coca Cola or Pepsi. KGBot also asks you questions to find out whether you need the password for the high-end S5720 HI, or for the low-end S5720 SI.
In addition, an enterprise knowledge cloud that supports device-cloud collaboration is the only way forward for smart customer service.
Explicit knowledge is knowledge expressed in written words, diagrams, and mathematical formulas, such as documents and PPT materials. Explicit knowledge can be likened to the tip of an iceberg above the sea. Implicit knowledge refers to knowledge that exists in the brain, such as skills, secret methods, intuition, and concepts of excellent employees. Implicit knowledge is like the huge iceberg that lies below the water, which is difficult to notice.
On the one hand, we construct knowledge graphs of explicit knowledge for knowledge structuring and reasoning. On the other hand, we need to build a platform to manage the implicit knowledge under the surface and gradually make it explicit. For example, the feedback from an external customer may be the root cause of another customer's problem. The problems found in an inspection may be an important source of KPI exception detection for the next time.
In addition, due to security issues, customer networks and data cannot be synchronized to the cloud for analysis and processing. However, without external data and features, customer problems may not be located. The device-cloud collaborative knowledge is the solution for this. Based on massive documentation knowledge, expert forum knowledge, and O&M knowledge, the cloud platform trains machine learning and deep learning models such as root cause location, disk detection, and KPI analysis. When customers conduct onsite fault location, they can go to the cloud to assist end-to-end testing in problem location. The effects of end-to-end testing are good for locating faults. On the one hand, knowledge can be desensitized and moved to the cloud after customer approval. On the other hand, the model can be adjusted, desensitized, and synchronized to the cloud for future fault location.
The TaskBot supports customer environment awareness, smart O&M, decision making, and resolves faults
The TaskBot is a customer service robot that performs smart O&M based on network environment and device status. It is also a semi-automatic tool for resolving customer faults, especially complex faults. Similar to L3 and L4 in self-driving, faults are classified by customers. Service availability faults can be rectified before reporting. However, if a service security fault occurs, the TaskBot confirms the fault and then rectifies it. According to Gartner's Intelligent O&M Analysis Report (2018), in the future, the proportion of enterprises utilizing smart O&M (AIOps) will reach 60%, and half of all faults will be automatically repaired. Therefore, using the TaskBot to realize fault self-recovery is the only way forward to automated customer service.
Faults, especially faults related to multi-device interconnection, are closely related to the customer network environment and device running status. In addition, analysis methods and diagnosis models vary with fault type. To resolve customer faults, the TaskBot needs the cloud bastion host and the probe to probe into the network environment and the context status of device operations. It also requires smart O&M algorithms and decision-making capabilities for different O&M scenarios. These are the most important features of the TaskBot. Similar to KGBot's device-cloud synergy, the TaskBot also requires device-cloud synchronization. Unlike the KGBot that returns only solutions, the TaskBot can run on the client and provide services independently, which is dubbed as "smart assistance" in the industry.
The TaskBot network agent, including the bastion host and probe, can obtain the networking and device run status while ensuring security. The networking status, fault environment, network cascading status, and the switch stacking status are all important for fault demarcation and root cause location. Similar to the coffee shop scenario, the system is unlikely to recommend soybean milk to you. The run status of devices is especially important in traffic prediction or sub-health detection. A sudden traffic surge at one point may be the root cause of network interruption in the next.
The TaskBot's O&M decision-making is another important component of customer self-service. The TaskBot's O&M decisions can be classified into four types: (1) routine maintenance, including log mining, alarm compression, and log analysis; (2) exception detection, including disk detection, KPI exception detection, and network sub-health detection; (3) prediction and prevention, including disk capacity prediction, performance prediction and capacity warning; (4) root cause analysis, including fault demarcation and location, fault diagnosis, and alarm root cause location. Time-Space decomposition is a common algorithm for log compression. KPI classification/clustering is a common algorithm for KPI exception detection. Time sequence prediction algorithms such as ARMIA are core algorithms for disk capacity prediction and warning. Collaborative filtering and association analysis are core algorithms for fault demarcation and locating. For the four scenarios, the TaskBot provides 20+ typical machine learning and deep learning algorithms to enable customers to analyze problems and perform self-service.
The challenge of ICT is how to build a secure and reliable customer service platform to provide efficient and reliable smart products and services
At the forum of State of Security Governance 2017- Where Do We Go Next of Gartner's Security and Risk Management Summit, analyst Marc Antoine Meunier delivered a speech, in which he compared data security governance to the "eye of a storm", emphasizing its vital importance in data security.
Secure and reliable smart products and services are also the digital twin requirements of "intelligence". How to define and implement data security governance in the smart customer service system, how to build data security and provide reliable services, and how to integrate into network security and ecosystems are the new challenges for customer self-service.
First of all, we need to understand that data security governance is not just a product-level solution that combines tools, but a complete chain that runs through the entire organizational architecture from the decision-making layer to the technical layer, and from the management system to tool support. Industry experts, third-party service providers, partners, customers, and organizations at all levels must reach a consensus on the objectives and purposes of data security governance and ensure that appropriate measures are taken to protect information resources in the most effective manner. This is also Gartner's basic definition of security and risk management.
The four important steps for data security governance include how to customize the data security governance process, including establishing the management accountability system and decision-making permissions, determining acceptable security risks, controlling security risks, and ensuring the risk control effectiveness. Data security governance must be a complete closed loop. Security assessment and specific indicator measurement must be performed to ensure that risks are effectively managed. Otherwise, the first step must be performed to correct the whole process.
Then, how do we tell good governance from bad governance? After setting objectives for the data security governance process, decision makers need to pay attention to several key indicators to determine whether the data security governance work is healthy and lighten the burden of enterprises. Gartner also provides several evaluation criteria for us.
Data security governance is also a long-term challenge for mutual, self-service, automated, and intelligent services.