Museum Friend

Motivation

This project was conducted at National Museum in Stockholm. I always want to learn more about collections while visiting museums. So it makes me think about a better method to interact and communicate with the artworks in museums.
In this project, I designed and developed a chatbot with chatGPT to help visitors learn more about the exhibition and collections.

User research

To discover how to improve the visiting experience by providing information on artworks, I conducted a research on how tourists learn in museums. The research combined online questionnaires and observational interviews.

From the questionnaire, I got a general understanding of visitors' information perception habits, their attitude towards interactive media in the museum, and their needs during the visit.

I selected 5 participants from the questionnaire responses. I invited them to take part in user interviews at the National Museum's Garden exhibition. I recorded their routes in the exhibition, and had interviews on their visiting experience.
‍

Pain points

Navigation

Before they start visiting, a layout of the exhibition can help them build an overall picture, so that visitors can get an expectation of the space.
In the in-between space, they always don’t know where to go next. They need a map in this situation.
‍‍

Description

For the participants who lack knowledge of the region or Greek mythology, even the descriptions cannot help them understand the artworks, due to some terminologies they don’t understand.
Some descriptions don’t contain information they are interested in. For example, besides objective knowledge, such as the genre, artists, and history, Jordan wants to learn something special from subjective aspects.
‍‍

Usability

It’s not easy to find the audio in the list and can be searched by names. Some participants gave up using it during the half visit.
‍‍

Voice

The voice of the audio is so calm that it makes participants easily get bored.

Ideation

Iteration

Use valid content‍

As we known, chatGPT sometimes makes up things, so does it in the museum tour guide. However, for the service in the museum, the delivered content must be accurate. So I update the way to generate content. I trained the model by the text from valid source:
  -  Wikimedia
  -  Museum websites
  -  Published books
‍‍

Choose preference‍

The feedback shows that some visitors only want text or audio messages. During the visiting, they would change the preference. As a result, I modified the message type to options, allowing the visitor to choose their preferred format.‍

Adjusting the tone‍

The role of the chatbot is more like a friend rather than a staff at museum that makes the visitor feel relax. Therefore, I changed the prompt for chatGPT that “You are visiting the museum with your friend” to adjust the tone more friendly.‍

Deliver

User test

I conducted the user test in the National Museum, and invited 30 visitors to explore the Garden exhibition. Half of them visited the exhibition with the chatbot, the other half with the Application developed by National Museum. After visiting, they filled a questionnaire about their using experience.‍

Result

In this project, I designed and developed the chatbot successfully. I had a great experience throughout the whole process. The user test shows that the chatbot assists visitors have an interactive experience and gain deeper understanding of the artworks.