Core demands
When the car owner has a car problem but does not know whether it is serious, or has no time to go to the 4S store for diagnosis, it is a very common demand for the car owner to get fast and accurate professional answers. Auto masters have provided more than 20 million auto question and answer services for car owners through matching professional technicians, and have developed into the largest auto question and answer platform and professional technician crowdsourcing platform in China. With the development of AI technology, it is possible to introduce intelligent Q&A into the field of high-frequency car Q&A. It is inevitable to adopt AI technology to further improve the efficiency of automobile question and answer service and user experience.
Solution
Auto masters took the lead in realizing intelligent Q&A in the tire field after accessing Baidu's understanding and interaction technology UNIT (customized human-computer dialogue platform). In combination with the rich question and answer knowledge base accumulated in the automotive field, automotive masters have successfully built an intelligent question and answer system in the automotive field on the UNIT platform using two skills: [question and answer pairs] and [multi round dialogue]. When the car owner asks relevant questions, the system naturally communicates with the user and presents the corresponding car solutions to the user. The specific solutions are as follows:
Flow Chart of Intelligent Dialogue Solution for Auto Masters
Auto masters have accumulated a knowledge base of thousands of pairs of questions and answers in the automotive field. It can be found that 90% of users' problems in the tire field focus on: tire burst, tire pressure, tire repair, spare tire, etc.
1. In order to solve 80% of the common fixed problems of users, we can solve them by matching [Q&A Pair]. After sorting out the historical question and answer service data in the tire field, a structured and clear question and answer pair is formed and imported into the UNIT platform to quickly configure the question and answer pair.
(1) Users can ask the same question in a variety of ways, and diversified questions need to be answered accurately, which can be achieved through the "many to one" matching of question answering units. For example, "how often to replace the tire" and "when to replace the tire better" both correspond to the answer of "tire replacement frequency";
(2) In order to make the robot's response more flexible and close to the real person dialogue, it can be realized through the "one to many" and "many to many" matching of question and answer units. For example, "Thank you" can get "You're welcome" and "I'm honored to help you" and other responses.
Example of Q&A pair matching method
2. In question and answer in the automotive field, in order to accurately locate user needs, it is usually necessary to obtain a variety of key information from users, such as car brand, model, parts, etc., in order to locate the problem feedback and correct solutions. This makes use of the multi round dialogue function of the UNIT platform. The auto master sorted out the common problems of users in the tire field: spare tire, tire repair, tire pressure, etc., configured detailed dialogue logic and trained models. Users can change various ways to ask questions. The question answering system can obtain users' intentions and necessary information through multiple rounds of dialogue and clarification of counter questions, so as to locate questions and reply to corresponding solutions.
Example of multi round dialogue
The realization of multi round dialogue depends on the configuration of dialogue units. The specific configuration process is as follows:
Step 1: Configure intention and word slot, that is, the user's purpose and key information to ask;
Step 2: Configure the script and trigger conditions for system response;
Step 3: Label the dialog sample. That is, give a demonstration to the dialogue system to teach it how to understand the intention, which word is the key information, and what the corresponding word slot is in the specific sentence that the user says;
Step 4: Training verification. After completing the dialog logic configuration and data annotation, the effective dialog ability can be triggered. The platform provides a dialog box that can directly experience the effect. When badcase appears, it can also be corrected in the dialog box.
Step 5: Optimize the effect. Robots, like humans, need a process of continuous improvement. It can continuously complete the configuration and labeling process in the platform to improve the robot's dialogue ability;
In addition, by integrating the "greeting" and "chatting" skills of UNIT, the system can naturally communicate with users when they ask questions unrelated to the car.
