Pocket Counselor

Midterm stress - modified response
Midterm stress - default response
Failed exam - modified response
Failed exam - default response

Inspiration

According to research statistics, 80% of students report that they feel stress on a daily basis and 40% of them don’t actively seek help. The response we get when we are seeking emotional support is often “Sorry, I don’t understand.” or “I’m sorry to hear that.” and this hasn’t changed for many years (at least from 2017). Because smartphones and its built-in voice assistant become prevalent these days, we want to create a pocket therapist that offers immediate emotional support whenever and wherever people are.

What it does

The Pocket Counselor project is an add-on to the default voice assistant software that provides more human responses to help students deal with stress at school. Imagine talking to Siri/Google Assistant about how stressful you are during midterms and instead of hearing “Sorry, I don’t understand.”, they reply to you with “Take some time and go over all the topics you don’t understand. You can do it!” We have trained the voice assistant to recognize and offer different responses to problems related to schoolwork, friends, family, suicidal tendencies, and negative emotions such as sadness, frustration, stress, and depression, etc.

How we built it

We built this prototype through the developer platform Actions on Google for Google Assistant. With hours of research into common problems students face, we put at least 5 example phrases as Machine Learning training data for the agent for each intent we have covered. The intents include SchoolPorblem, FamilyProblem, JobProbelm, RelationshipProblem, ImposterSyndrome, LifeSaver, Encouragement, Appreciation, etc. We also provided at least 5 example responses the agent can reply to each intent. During our testing stage, we are able to do live demo through any phones with Google Assistant installed and developer accounted linked.

Challenges we ran into

Feasibility: Our original idea is to make the voice assistant recognize whether user has potential symptoms of mental illness and then provide real-time assistance 24/7. If a user demonstrates severe mental stress or depression, the voice assistant will then seamlessly connect a counselor with the user during the conversion. However, this idea is hard to achieve in reality. Since this service is available for users at any time, it might be difficult to connect users with a counselor at midnight, which involves certain operational costs that cannot be avoided.
Training data: Currently, we manually typed in possible conversations between a student with potential mental illness symptoms and the voice assistant. Since it is not a comprehensive dataset, there are use cases that we did not include. For this to be said, our current outcome provides a solid framework that combines counseling feature with voice assistant.
User Utterance: How to understand user utterance in a broader context of the dialogue is the main challenge while developing the add-on counseling feature of Google Assistant. In Google Assistant’s developer platform, we can add parameters such as dates, times, and places that define which parts of the user utterances to extract. However, Google Assistant does not allow us to create parameters related to user’s emotions such as sadness, anger, and depression, which can largely improve its ability to identify user’s intentions behind his or her move (e.g. asking a question or making a complaint).

Accomplishments that we're proud of

Built human-like voice assistant with better user experience
Created potential benefits to students and their communities
Learned research statistics related to mental health and how technology can help

What we learned

Prototyping is fun, although we faced many problems. We’ve applied voice technology, Machine Learning, and Natural Language Processing/Understanding (NLP/NLU) in this 24-hour project.

What's next for Pocket Counselor

More intents to be covered, including discrimination against race, sex, sexual orientation, ethnicity, etc.
Find ways to integrate the project with the built-in Google Assistant training data