Duration:
One Semester
Lab:
HCI/IC/EPFL
Goals:
Build a conversational chatbot that generates questions with emotionally appropriate intents
Assistant:
Ekaterina Svikhnushina (ekaterina DOT svikhnushina AT epfl DOT ch)
Student Name:
Open
Keywords:
Question generation; conversational chatbots; natural language processing
Abstract:
Questions constitute over 60% of human conversations and appear a key conversation driver. Yet, only few studies exist that focus on the development of conversational chatbots capable of asking meaningful questions (W. Wang et al., 2019; Y. Wang et al., 2018). The aim of this project is to take the next step towards automatic generation of appropriate and attentive questions leveraging the taxonomy of question types and intents which was recently established by the lab. The resulting model is expected to ask suitable questions given the emotional context of the dialog to approximate emotional regulation strategies occurring in human-human conversations.
The student is expected to:
- Survey the related work on conversational question generation, taking note of employed methodological approaches and datasets.
- Build a question-generation model taking into account labeled question types and intents from the EmpatheticDialogues dataset (Rashkin et al., 2019).
- Evaluate the model performance with automatic metrics and human judgment.
Related Skills:
Ability to work independently; strong knowledge and skills in natural language processing and neural networks; strong analytical skills; knowledge of reinforcement learning is a plus.
Suitable for:
Master student. Interested student should contact Ekaterina Svikhnushina (ekaterina DOT svikhnushina AT epfl DOT ch) and Pearl Pu (pearl DOT pu AT epfl DOT ch) along with a copy of your CV.
Duration:
One semester
Goals:
Develop guidelines for designing a crowdsourcing task for textual emotion annotation.
Supervisor:
Dr. Pearl Pu
Student Name:
Open
Keywords:
Emotion Annotation, Crowdsourcing, Design Guidelines, Design Recommendations, Task Design
Abstract:
Annotation of emotions in collected text documents is an essential step for developing the reliable emotion classification system. It can serve as the ground-truth for training and/or testing emotion recognition models. Current crowdsourcing approaches involve annotation of text documents for emotions and emotion indicators. Yet, subjectivity in understanding of the emotion concept and lack of clear instructions may result in high inconsistency of collected data. There is a need to elaborate the appropriate design guidelines in order to ensure the collection high-quality annotations.
The responsible student will:
- Survey the existent approaches to crowdsourcing sentiment and emotion annotation of text, as well as the method of their qualitative and quantitative evaluation.
- Analyze data collected from the previous crowdsourcing experiments on emotion annotation.
- Based on that analysis, derive design guidelines for crowdsourcing tasks on textual emotion annotation.
Related Skills:
Strong analytical skills; Statistical analysis; Interest in Crowdsourcing, Qualitative Research, and Human-Computer Interaction Design
Suitable for:
Bachelor or Master Student
Duration:
One semester (can be adapted to both bachelor and master students)
Goals:
Improve our prediction capability of influence among reviews from e-Commerce websites, and our understanding of human behavior in this setting.
Supervisor:
Dr. Pearl Pu
Student Name:
Open
Keywords:
emotion recognition, influence detection, prediction, social media analysis, natural language processing
Abstract:
Reviews keep playing an increasingly important role in the decision process of buying products and booking hotels. However, the large amount of available information can be confusing to users. A more succinct interface, gathering only the most helpful reviews, can reduce information processing time and save effort. To create such an interface in real time, we need reliable prediction algorithms to classify and predict new reviews which have not been voted yet but are potentially helpful. So far such helpfulness prediction algorithms have benefited from structural aspects, such as the length of review or its readability score. Since emotional words are at the heart of our written communication and are powerful to trigger listeners’ attention, we believe that they can serve as important parameters for predicting helpfulness of review texts. This is an excellent opportunity to develop a combined approach of Machine Learning and Natural Language Processing to improve the prediction of these influential reviews and provide a light interface to users.
The responsible student will:
- Familiarize with existing techniques in influence prediction and emotion extraction
- Discover and try the existing framework for prediction
- Develop new learning mechanisms to extract emotions from review texts
- Understand the features of interest in the prediction process
Related Skills:
Programming skills (Python, Matlab or similar for prediction); Interest in Machine Learning, Natural Language Processing, and/or Text Mining; French or English mother tongue a plus
Duration:
One Semester
Lab:
HCI/IC/EPFL
Goals:
The aim of this project is to provide NI (natural interface) or mixed interface to BIM visualization tools.
Supervisor:
Dr. Pearl Pu (pearl DOT pu AT epfl DOT ch)
Keywords:
voice and dialog interface for CAD tools
Abstract:
BIM visualization tools allow users to walk through a house or a building in a virtual or augmented digital world. However, the current tool, using mainly graphical user interfaces (GUI), is difficult for most lay users. Replacing it with voice and dialog interfaces (natural interfaces) or integrating them with GUI can potentially overcome this challenge.
The student is expected to:
- Investigate and use some existing BIM visualization tools
- Integrate DialogFlow (goole’s dialog engine) to GUI
- Integrate voice activation
Related Skills:
HCI; basic knowledge in voice and dialog interface, e.g., DialogFlow (Google)
Suitable for:
Master Student. Interested student should contact Pearl Pu (pearl DOT pu AT epfl DOT ch) along with a copy of your CV.