Alibaba Xiaomi: Applying Speech Recognition, Semantic Analysis and Deep Learning on Taobao Mobile
Created#More Posted time:May 9, 2016 18:10 PM
The development of artificial intelligence over the past 60 years has led down many winding roads, but has ultimately borne fruit. You may recall the analyses and prospects laid out in the 2015 report "Academician Tan Tieniu: Properly Managed, Artificial Intelligence is a Blessing, Not a Curse". This year, the man vs. machine
Go competition has made artificial intelligence a hot topic.Brilliant work is being done in academia, industry, and research institutions.Everyone in the field is hoping to learn about Alibaba's efforts in artificial intelligence technology. We have extended a special invitation to the Alibaba technology expert Kong Wu to discuss the Mobile Taobao client's speech recognition, semantic analysis, deep learning, and other AI technologies used to create the Alibaba Xiaomi customer service system.
As the competition between man and machine has attracted worldwide attention to artificial intelligence, some people are beginning to wring their hands over humanity's future.
Competitions between humans and machines draw great public attention, but the value of artificial intelligence goes far beyond the competitions. In fact, a great deal of artificial intelligence has already become part of our daily lives, profoundly influencing and changing us.
Currently, artificial intelligence is being commercialized well beyond Silicon Valley. BAT is also exploring the area deeply. For example, AI is transforming the service industry due to its speech recognition capabilities.
Many people easily equate answering a phone with customer service. However, nine years ago, most people thought cell phones could only be used to make calls and text messages. Then how would people understand customer service in five years?
Business Needs Drive Alibaba Toward Intelligent Customer Service
As the world's largest e-commerce platform, Alibaba copes with over 400 million active consumers, millions of active sellers, thousands of Alibaba products and services, and millions of transactions each day.
According to the magazine Tianxiawangshang, Taobao and Tmall receive nearly 50,000 calls to the customer service hotline each day. The online service traffic from wireless terminals is even more staggering, with more than 1 million service requests a day. Moreover, the figure is still growing rapidly. Today, young people prefer to seek help online via cellphones.
The complexity and surge of Alibaba Group's business has resulted in an increasingly "bloated" customer service staff. Given the explosion in customer demand, simply increasing the team of customer service staff and raising efficiency is obviously not a long-term solution. Moreover, Alibaba does not want to become a predominantly customer service-oriented Internet company.
Obviously, Alibaba already has one of the world's most complex customer service systems. Therefore, intelligent customer service is the only path to sustainable development.
How to Deliver a "One-Click" and "One-Stop" Service Experience
Putting users through or handling their transaction disputes is far from enough. Consumers and sellers both need a "One-Click" experience. Taobao and Tmall have naturally closed transaction loops, as well as extremely rich data and advanced infrastructures. In this situation, how can we use technology to provide users with a "One-Click" and "One-Stop" service experience?
Starting in 2015, Alibaba released a next-gen intelligent customer service product for the Taobao mobile client: Alibaba Xiaomi. This product is powered by a range of AI technologies such as speech recognition, semantic understanding, personalized recommendations, and deep learning. In response to millions of service requests each day, this intelligent solution can handle up to 80% of requests which exceeds the industry average for intelligent customer service products by more than 60%. In some key scenarios, this rate can even
reach as high as 95%. In the service field, the exact match rate of semantic intention in human machine conversation has been raised to 93%, doubling the rate of customer satisfaction for traditional self-service systems.
Unlike Apple's Siri and Microsoft's Xiaoice, Xiaomi embodies Alibaba's new insights about AI. The most useful aspect of AI is its ability to carry out in-depth human interactions. Specifically, it simulates human ways of thinking in order to serve real people. To achieve this goal, AI must be built on a great deal of data and scenarios. Each day, Alibaba produces a massive volume of transactions, and the user inquiries about each transaction make Xiaomi increasingly smarter, so it is highly feasible to put Xiaomi into commercial use.
From Intelligent Customer Service to an Intelligent Assistant, Xiaomi Can Do More
Benefiting from the growing big data base, we can proactively analyze and predict consumer service needs and reach out to users. Thus, Alibaba Xiaomi will reduce the need for users to seek assistance from human customer service staff via phone calls or online chat by 70%. By analyzing and predicting massive numbers of questions during the consumer inquiry peak on November 11 each year, the service team's algorithm engineers can have the knowledge base automatically updated in minutes for the first time in the industry, and predict the problems each user may encounter in real time. Before users have even asked for help, we can send them precise service information to greatly ease their concerns.
Unlike traditional intelligent customer service products that are based on a custom knowledge base of questions and answers, Alibaba Xiaomi combines AI and
Knowledge Mapping to support contextual understanding in multi-exchange dialogs and personalized memory. Additionally, the robots will learn from millions of
manual customer service records and a massive volume of knowledge sources, helping them automatically improve their intelligent problem-sovling
capabilities. A group of speech recognition scientists have enabled Xiaomi to understand users' issues in a real-time conversation. Using the world-leading
BLSTM algorithm, they have created a bidirectional, long-term/short-term memory neural network. By breaking through the bottleneck of traditional algorithms only capable of processing data from left to right, the new approach allows for recognition in both directions simultaneously. Bolstered by Alibaba Cloud's powerful cloud computing capabilities, Alibaba was the first to deploy the BLSTM algorithm for a large enterprise to effectively shorten the latency caused by bidirectional recognition. This has significantly improved the accuracy of speech recognition. This is especially effective for users with strong accents, who speak fast, or who are in a noisy environment. Alibaba Xiaomi has already evolved from an intelligent customer service product into an intelligent assistant, like Apple's Siri or Microsoft's Cortana. Alibaba Xiaomi can even joke with users. It can help them solve difficult problems, or assist them in recharging their phone accounts, checking the weather, buying plane or movie tickets, buying flowers, and doing more. Thanks to the in-depth
application of AI technology, customer service is moving from the backend to the frontend, helping other business lines improve user experience.
The "Honeycomb" Intelligent Decision System
In the future, Alibaba Xiaomi will try to help users solve problems such as: What gift should I give on Women's Day? How should I choose a smart TV? And
where should I go this weekend ?
With the increasing maturity of smart robot service, Alibaba will make the service available to sellers operating on its platforms. Sellers will be able to use simple configurations to provide consumers with exclusive and convenient services. This will eliminate the need for many smaller sellers to attend their computers 24/7, since Alibaba smart robots can help users answer many simple questions.
When a robot cannot solve a problem, Xiaomi's smart routing system will seamlessly refer the user to a real customer service representative based on the "experience first" principle.
To fundamentally increase the efficiency and user experience of the human customer service workbench, Alibaba's customer experience business division's
technical research and development team is developing a "Honeycomb" intelligent conversation assistance system. Honeycomb will analyze all the actions of users before they turn to customer service, completely integrate user statuses and order information, and dynamically analyze each conversation between users and customer service to provide precise reply recommendations in the customer service input box within 10 milliseconds. The customer service representative only needs to select and modify the appropriate response. This will help maximize the work efficiency of the customer service staff. Honeycomb's constantly expanding smart plugin removes the need for customer service staff to jump out of the system, even if a user suddenly asks, "What is the weather like today?" or What is BABA's stock
The conversations between customer service staff and users will become the raw data for machine learning and be used to optimize this intelligent conversation
assistance system. After user conversation habits and customer service staff' personal styles build up for a period of time and learned by Honeycomb, the
system will become accurate and personalized so users will not feel they are simply being fed stock answers.
Transaction dispute processing is much more complex than answering user questions. Taobao and Tmall are e-commerce trading platforms where a huge number of buyers and sellers do business. This can occasionally lead to problems, or in the worst cases, disputes. Once a dispute arises between a buyer and a seller, Alibaba customer service is there to solve problems and resolve the dispute.
To handle the most complex aftersales transaction disputes, Alibaba's data scientists have begun to attack the "last bastion" of customer service by launching a secret project with the internal codename "Wali". Because transaction disputes involve the different demands of buyers, sellers, and logistics companies, a dispute may involve more than 50 dispute adjudication factors, making the task extremely complex. In the past, only expert customer service representatives that had received a long period of training could correctly arbitrate in disputes. Moreover, businesses hesitated to use machine algorithms to tackle problems that required so much careful consideration and complex operations. However, with the evolution of intelligent technology and the collection of data to reflect precise behaviors, an algorithm deployed on a large-scale distributed computing cluster can study millions of case judgements within five minutes. Through continuous optimization and calibration, machines
will be able to accurately resolve complex disputes just like a human.
With a focus on communication and intelligence in the modern customer service field, the model that combines channel, CRM and human customer service will
change. Using mobile and IoT technology, the service contact point will no longer be a single phone. Instead, we will see the development of a user-centric, experience-driven, comprehensive service model that is open to users anytime and anywhere. In addition, intelligent voice conversation systems based on speech recognition, semantic understanding, and other technology will extend the capabilities and boundaries of customer service. Without specific customer service equipment, services can be closely linked with the business. Thus, customer service will be thoroughly redefined.
The article was taken from the Alibaba Cloud Yunqi Community website. If reprinted, retain the author and source (Alibaba Cloud Yunqi Community) and send an email notification to Yunqi at (email@example.com).
1st Reply#Posted time:Jun 8, 2016 7:14 AM
Is Xiaomi providing voice recognition capabilities to Alibaba's support group? If so, that's interesting because I didn't know Xiaomi offered that.
Can you point me to a URL that describes these offerings from Xiaomi?
I ask because I'm writing a story on Xiaomi's entry into the drone business and this is a good background story about their capabilities in AI and deep learning.
The Robot Report