Human-like learning dialogue robot in the use of Magic – Sohu Technology fifa14下载

Human-like learning in the use of magic in robot dialogue – Dialogue robot technology Sohu like Siri, a lot of Peru, wheatgrass, Allo can have time with you in your poor garrulous, but with the manufacturers and users to make sense of a highly versatile dialogue robot is not realistic, dialogue robot the attitude is changing — manufacturers have subtle attempts to start with some vertical deep roots, and from pure chat robot development to the dialogue what tasks can be completed for the specified user function. All of a sudden "dialogue robot talk" to find a new scene. However, this new upgrade of the dialogue robot, the focus shifted to replace people to make decisions, and help users to complete the task. How does it understand what the user wants to do? How to do the chat process can not be achieved in the Human leaering? And then how does it know that the third party App can complete the task, which needs to click on, and then can be done with the user wants to do things butt? Like What is its core technical requirement when it becomes a task for the user? …… These only have many years of practical experience will have the answer, we invited the founder of Dai Shuai Xiang Mu ran cognition, small home to get him for the robot and "chewing" dialogue process intention, realize human decision finally, help the task execution. Dai shuaixiang, former director of Baidu architect, had long served as Baidu Query to understand the direction of the person in charge, is the natural language understanding technology experts, who won the first and so far the only one to NLP technology as the core of the Baidu highest award. 2010 proposed Query rewrite model to Baidu search engine technology has brought a leap across the times, making the search for relevance and advertising revenues are significantly improved. The model is similar to the academic model of the "Query Rewriting Monolingual Statistical Machine Translation" was put forward nearly 1 years earlier. This model has so far been widely used in all Baidu search product line. Application of Natural Language Processing, semantic search, automatic problem solving and other fields more than 20 patents. First of all, what is the Human-like learning highlighted in your product? One-shot and RL in the dialogue model (Learning). One-shot Learning was designed to carry out the training from small samples, in order to solve the dialogue system of "cold start" problem; RL is an unsupervised learning, it is through trial and error approach to learning. In view of the cold start dialogue model, RL can help the system to enhance the dominant strategy in the dialogue model and weaken the influence of the negative strategy. The actual performance is that users will feel more and more user-friendly system, or personalized. The above two learning methods.相关的主题文章: