Projects

Community Building Projects

REQUIREMENTS:

• Software development (medium)

• Statistics (medium)

• Web design (medium)

• Community building (high)

• Level of difficulty (low to intermediate)

Crowd-sourcing recognizing animals

Similar to the Merlin smartphone app, which recognizes birds by their shape and song, develop an app and build a user community that helps collecting labeled pictures of dogs, horses, but could also be insects (e.g. ants, bees, butterflies), and recognizes them by pointing the camera to the animals. It could either be the breed/species of the animal, or the emotion for an animal where humans can guess their emotions (e.g. dogs, horses, cows).

This project builds on previous work of a COIN course team (goodboi.camera, paper) developing a responsive app that detects the emotions of dogs by pictures. A possible extension would be to buid a native smartphone app, or exted it to horses, reading the facial emotion and body posture of dogs and horses, developed in previous COINs seminars, to build a system on Android or iPhone that labels the emotions of dogs, horses, and other animals that you look at, based on their body language. You will have to collect libraries of labeled pictures of e.g. dogs or horses, where experts will have to agree on the emotion of the animal. For instance a wagging tail of a dog is correlated with joy. This is an active field of research.

Coolhunting Projects

REQUIREMENTS:

• Software development (medium)

• Statistics (medium)

• Web design (medium)

• Community building (low)

• Level of difficulty (low to intermediate)

Measuring Covid Depression

Analyzing the last 5 years of the COIN course email data, measure changes in emotions and values of the participants over the last 8 to 10 courses using NLP emotion recogntion, and the other personality characteristics modules available in Griffin. Analysis could also be extended to try to find other changes in network structure, emotions, and other socidemographic variables.

Protest analytics on social media

Protest movements like the Black Lives Matter Movement are portrayed in various ways on social media which influences a user's perception thereof. This project is comprised of two tasks: First, create a classification algorithm with image and textual data to categorize the way protest movements are portrayed on social media. Second, analyze user reactions according to the classification results and detect if user opinions of the protest movement change over time. Are there any opinion leaders who might influence the general perception of the movement and how do their messages diffuse within the social media discourse?

Ethical algorithms

This topic revolves around experimental projects focusing on how AI-based technologies can foster ethics. There are two different directions of experimental research: On the one hand, students can explore whether text- or voice-based digital assistants are able to drive ethical behavior of users (for instance by building a chatbot that is offering product references of more sustainable alternatives to influence the users shopping decision). On the other hand, students can explore criteria of implementing algorithm ethically (for instance by first comparing AI-based performance evaluation systems in the context of call center applications with those of a ballet dancer applications; and second by comparing the effects on how the applicants assess the judgment abilities of the algorithm in comparison to humans depending on the final decision of the system).

Authenticity of corporate influencers

Many organizations (especially SMEs) struggle with their social media performance. Since some of their employees are highly influential on social media, corporate influencers are one way to increase an organization’s visibility on social media. However, not every corporate influencer contributes to a better perception of the organization since there is a lack of authenticity. Hence, in this project, students should explore indicators of successful corporate influencers and the role of authenticity.

Prediction Projects

REQUIREMENTS:

• Software development (high)

• Statistics/machine learning (high)

• Web design (medium)

• Community building (low)

• Level of difficulty (high)

Measuring Human Emotions with the Happimeter, Face Emotion Recognition, and Plants as biosensors

This is an extension of previous projects in earlier COIN seminars. The basic idea is to us AI to measure human emotions through body sensing technologies such as the happimeter, face emotion recognition, voice emotion recognition, and plants as biosensors. Possible ideas are:

Using previously developed emotion tracking systems, such as the happimeter, facial emotion recognition in zoom meetings, and plants as biosensors, measure the emotions of student participants in the Jazzaar festival in Aarau, March/April 2022. Collect emotion data from students during their rehearsals in Aarau, and from their teachers, world famous musicians, who this year will join the students by zoom. Success of each lesson will be measured by a brief survey. This is an extension of different projects from previous COIN seminars.

In a project jointly with Maison Du Futur, a collaborative of performing artists, develop a system that will give augmented reality feedback to the observer of a performance based on the brainwaves of the observer. In an experiment in winter 2021/22 we used the Emotiv Brainwear Insight headset to read the brainwaves while observing the performance, we will now extend it with the plant spiker box and face emotion recognition by measuring the action potential of Mimosa Pudica and other houseplants such as basil to track human-plant interaction using the spiker box and other hardware and compare it with the human alpha brain waves measured with the heart and brain spikerbox.

Another possible project would be the development of a plant sensing system (e.g. using Arduino and Raspberri Pi) using electric sensors which is capable of tracking the perception of plants and their communication with their environment measuring their signalling in their chemical and physical communication gateways, extending the plant spiker box. This is an extension of a previous COIN seminar project building an action potential measuring system for plants.

Emotion detection from text by analyzing facial expressions of actors in TV soaps

This is a continuation of an ambitious project in last semester's COINs seminar, where two teams tried to compare emotion recognition from facial expression with emotion detection from text. The teams analyzed facial expressions of actors in different series to identify the emotions shown. They used these emotions as labels for the spoken text. Based on this data, we train several state of the art models to automatically classify emotions from text. Building models with LSTM and BERT embeddings the teams found that their approach performsd significantly worse than other stand-alone "emotions in text" label methods. This raises tantalizing research questions about the accuracy of human labellers to detect emotions in text. The goal of this project is to further investigate how facial and textual emotions can be accurately recognized through deep learning.

Clustering of text and image data from career counselling clients

In a joint project with ZHAW we aim to analyze data from over 1000 career counselling clients. Clients rated 80 images (appeals to me, neutral, does not appeal to me) and selected their favorite image. Finally, they wrote a personal story about their favorite image. Can we find a relationship between personality characteristics, selection of images, and personal story, by clustering the text data (personal story) and the selected images (content, colors, saturation, ...)? and comparing it with educational level, management level, gender, age as well as personality aspects.