Vinner: Enjoy your dinner with new friends virtually
Vinner is a service that arranges blind lunch dates and generates conversation topics in real time.
Background
COVID-19 has disrupted many social structures and norms regarding how people meet and communicate with each other. As a result of the pandemic, non-face-to-face communications have become increasingly common. Consequently, the use of online conferencing platforms such as Zoom, Google Meet, and Clubhouse has seen a significant rise.Problem Definition
However, in such an online environment, users often struggle to interpret non-verbal expressions and gestures. Moreover, sharing real-time generated reactions with others poses a challenge. Despite a general increase in social media usage, social interaction during the pandemic has significantly diminished.
Based on the data, we determined the necessary AI model: clustering of similar groups of people for conversations, accompanied by a real-time audio-to-text engine and natural language processing for question generation.
Interaction: The AI would analyze the commonalities between two people and then generate relevant conversation topics by reviewing previous discussions and users' preferences towards them. The human user could either accept the suggested topic or ignore it, and then provide ratings for the prompts after the conversation. This process results in an interaction where the AI recommends topics, and the human user either confirms or declines them.
Prototype
The Vinner service offers an AI-infused virtual meal conversation aid. Initially, the application collects information from the user to find a partner with shared interests. Upon successful matching, a notification is sent to the users, and a virtual meal conversation begins. During this virtual conversation, AI-recommended questions are displayed to guide the discussion. After the conversation concludes, a feedback page is presented to the users for their input. Further details are provided below. Access to the prototype is available here.
Matching Process
Vinner collects basic information to match users with suitable conversation partners who share similar interests. This information not only aids in finding a compatible partner but is also utilized to generate questions during the conversation.
The underlying principle is that individuals with common interests have more to discuss and are more likely to enjoy conversations centered around topics relevant to their shared interests.
AI-Recommended Questions
The main feature of the Vinner app is the AI-recommended questions or conversation topics, which aim to reduce awkwardness during conversations. The app uses pre-surveyed data about users' interests and their real-time conversation data to generate these questions.
By employing a speech-to-text API and a natural language processing API, Vinner extracts keywords from users' conversations and classifies their topics. With this processed data, Vinner's AI retrieves related questions from the question database and then filters them to consider ethical issues.
User Feedback
After the conversation ends, Vinner collects feedback from users. They are asked to complete a survey indicating whether they enjoyed the Vinner service and what aspects they appreciated.
Additionally, users rate their conversation partners. These feedback results are used to calculate the ratings of users that are displayed during the matching process.
The brand experience Vinner aims to create is 'friendly.' Branding plays a crucial role in shaping the perception of artificial intelligence's role within the system. Given that AI is positioned as the creator of conversation topics, we have intentionally crafted an image that is particularly friendly, akin to a friend, and trustworthy.
User Testing
During inter-group prototype testing, we received the following feedback from another team: They argued that randomly recommended questions by the system could disrupt the conversation. Even if these are AI-generated questions tailored to the conversation's context, the timing was often deemed inappropriate from their perspective. Additionally, there were concerns regarding potential misuse. Based on our observations, we noted that the timing of recommendations plays a crucial role in preventing disruptions in the conversation.Challenges and Future Work
However, we recognize a potential drawback that could lead to an increase in virtual interactions, potentially backfiring. There's a concern that some users may exploit this project for unintended purposes. Additionally, we have identified the risk of conversation disruption caused by the recommendation process as a significant challenge.Building on the achievements of Vinner, we have outlined our future endeavors. Firstly, we plan to enhance our approach by expanding user data collection and conducting beta testing. Secondly, we aim to develop an AI engine capable of real-time audio transcription, keyword extraction, and enriching our question/prompt bank. Thirdly, we have considered potential exit strategies, such as establishing a start-up or achieving recognition through design awards.