Source: Deep Learning on Medium
For the last three years Amazon has been testing the capabilities of the voice platform known as Alexa. The Alexa Prize Socialbot Grand Challenge is a competition for student teams dedicated to the development of conversational intelligence. This year Amazon received more than 375 applications from 22 countries and selected the top 10 teams according to the following criteria:
- Potential scientific contribution in the field.
- Technical value of the proposed approach.
- Novelty of the idea.
- Ability of the team to fulfill the submitted plan.
The Alexa Prize Socialbot Grand Challenge is a part of Amazon’s mission to make the voice assistant smarter, talkative so it can become useful and exciting for users. But Amazon is not the only company in this business: Google, Apple, and Samsung are also working on their assistants.
In the contest, each team competing for the Alexa Prize had an access to the Amazon’s automatic speech recognition service to get a textual interpretation of what users say, and their text-to-speech technology to provide a response. Our main challenge is to develop a social bot that will allow Alexa to conduct a conscious conversation on a wide range of topics.
In order to activate Alexa,it is necessary to vocalize the activation phrase “Alexa”. Alexa allows you to make to-do lists, record voice messages, warn you about bad weather, or play music that suits your mood. However, all these inquiries can quickly lead to a conversation about plans for the weekend. And our mission is to ensure that Alexa can address all these inquiries. In order to do so, Alexa needs to know the user preferences and combine several conversational skills. Sounds easy, right? In fact, this task is not trivial at all.
“You can think of DREAM Socialbot as an interactive and conversational gateway that stands between the user and a ton of online content,” says Yuri Kuratov, the leader of the DREAM Team. “However, the bot should use the two-way strategy focused on the user needs and the available content.
The Socialbot has to be sensitive to the users requests. Does the user feel positive about what Socialbot said? Or does the user still have negative emotions, and Socialbot should change the topic? And who is user? And what topics are interesting to him or her?
In the contest, the team dealt with:
- Infrastructure tasks: To work, the bot has to be deployed and tested somewhere, and it is also necessary to load and run programs and models.
- Research tasks are the most creative part of the whole process: It is necessary to come up with new models that will be integrated into the dialogue with the user. The full cycle ranges from finding a problem to formulating a research problem, collecting data, or searching for existing ones, creating a baseline model or improving existing models, and comparing them by metrics.
- Business tasks include the work directed to interest the user during a conversation.
- Data collection tasks — parsing news sites, more interesting and popular news, movies, ratings, and reviews on them.