No abstract

In multi-round dialogue systems, we can easily find that the final reply is closely related to two points, one is the context of the dialogue, the other is the persona characteristics. But not all characters and contextual information will affect the final reply, because the final reply may only be related to some crucial characters and contextual information,the indiscriminate use of all information may even have a negative impact on the generated dialogue. So it is necessary to extract and utilize key characters and contextual information to improve the quality of the final generated response. In this paper, we show how to solve this problem through our new model and methods. Specifically, our new model consists of two parts: encoder and decoder. The encoder is mainly used to encode personas, contexts and historical responses, and the decoder generates corresponding words from the vocabulary. Then, the weight of the character and context is updated through the multi-head self-attention mechanism to affect the response generated by the decoder. The experimental results show that compared with the baseline models, our model and methods have improved in terms of metric-based evaluation.

With the development of the times, the demand for executing programs that require massively parallel computing on embedded devices has become increasingly strong. To meet this demand, various heterogeneous systems have been built on SoCs. In order to better use and manage heterogeneous systems, and to facilitate the development of heterogeneous computing programs. People will implement a heterogeneous computing framework in the operating system. Unfortunately, the Reworks real-time operating system does not provide us with a heterogeneous computing framework for writing heterogeneous computing programs. We chose OpenCL as the heterogeneous computing framework for the Reworks operating system and implemented the kernel state of the OpenCL heterogeneous computing framework, also known as the OpenCL driver.

Intent recognition is the first part which needs to be accurately recognized in the conversation between ai and the user. Due to the diversity of user intent, it is difficult to manually define all categories of intent in advance during the training phase. The inability to distinguish undefined intent categories can cause the ai to keep replying to the user with seriously wrong answers, which can greatly reduce the user's sense of experience. Therefore, it is very necessary to realize the recognition of seen intention categories and the distinction of unseen intention categories. In this paper, a model with Automatic probability threshold which based on the bert model is used to ensure accurate recognition of seen intentions while distinguishing unseen intentions. Due to the inadequacy of training samples, the automatic probability threshold model in the paper uses text vectors from two bert models containing different dropout parameters as the training set, which can improve accuracy of the model.