Facebook Researchers Builds New Dataset for Training Personalized Dialogue Agents

0
103
Dialogue Agents

Facebook Researchers have reportedly collected and complied a dataset of 700 million persona-based and 5 million personas dialogue with an aim to enhance the end-to-end dialogue systems and for Training Personalized Dialogue Agents. This is likely to engage more dialogues between humans and computer agents.

Dialogue systems which are also known as conversational dialogue agents are designed computer systems which helps in communicating with human being through graphics, speech, text, and other methods, in a conspicuous way. As of now, dialogue system which is based on neural architectures like LSTMs or memory networks have been delivering fluent communication, especially when trained directly on dialogue logs.

In a recent study, a team of researchers at Montreal Institute for Learning Algorithms in collaboration with Facebook AI has created PERSONA-CHAT. It is a dataset which consists of dialogues between personas or agents with text profiles attached to them. The team discovered that training a dialogue system, particularly on persona enhanced their engagement in interactions.

In order to address the restrictions of the previous collected dataset, a team of Facebook researchers created a large-scale based on persona dialogue dataset. This dataset consists of conversations collected from Reddit, an online platform. In order to evaluate its effectiveness, the researcher tried to train person based end to end dialogue system on their brand new developed dataset. It was noticed that the systems which were trained on their dataset were capable of conducting more engaging conversations. This actually outperformed other dialogue agents. It is expected that in the future these findings might lead to in development of engaging chatbots.

Leave a Reply