Done Past Project

Modeling Effects Of Modifiers on Emotional Statements

Duration: Two semesters (February – June 2015, February – June 2016)
Goals: Investigate the effects of different linguistic modifiers on emotional expressions, and suggest how to model those effects within emotion recognition system.
Responsible TA: Valentina Sintsova and Pearl Pu
Student Names: Margarita Bolívar Jiménez (2015) and Nataniel Hofer (2016)
Keywords: Emotion Recognition, Text Classification, Social Media Analysis, Modifiers Effects

People express their emotions and feelings in multiple subtle ways. Even when they use explicit emotional terms, such as “happy” or “sad”, the emotional meaning of statements can change because of the variety of linguistic modifiers. Those include negation, intensity shifting, modality, and others. So far the researchers have investigated the effects of those modifiers on polarity of terms (positive vs. negative). However, their effects on more fine-grained emotion categories remain understudied. The first part of this project investigates the effects of different modifiers on emotional meaning of the terms via data analysis techniques. The second part studies to what extent the better modeling of modifiers improves emotion classification quality.

Advancing Human Computation for Emotion Detection

Duration: One semester (February – June 2015)
Goals: Design a human computation task that would allow collecting affective knowledge of better quality, and develop the evaluation techniques to quantify the impact of the task design.
Responsible TA: Valentina Sintsova and Pearl Pu
Student Name: Séphora Madjiheurem
Keywords: Human Computation, Emotion Recognition, Crowdsourcing, Amazon Mechanical Turk, Experiment Design, Quantitative Evaluation Techniques

Social media are filled with emotional content, which many researchers and companies seek to analyze. However, automatic methods for emotion recognition are far from the level of human ability to understand emotion language. Human computation techniques are seen as a way to help machines learn how to detect emotions. Online labor platforms such as Amazon Mechanical Turk allow to use individual humans to obtain answers to such judgment tasks as emotion detection in text. One strategy to obtain quality answers is to combine answers from different workers. Yet, in order to make use of the wisdom of the crowd, human answers must be comparable. This can be achieved by providing clear instructions and designing tutorials for the task. Moreover, if the human computation task is subject to systematic bias, using multiple workers is not enough to obtain quality answers. In this project, two experiments were conducted in an online labor platform. The first experiment aimed to evaluate the impact of tutorials on the quality of the answers provided by workers and on their engagement in the task. The second experiment was focused on comparing the workers’ output quality and engagement when using different incentives for motivating workers. The results show that tutorials with limited instruction do not necessarily lead to poorer performance. The results also demonstrate better quality work from workers under certain treatment conditions for motivation depending on the difficulty of the task.

#Travel: Search for topics shared among city locations

#Travel Landing Page

Title: Automatic detection of topics in the city and the associated locations
Duration: One semester (September – December 2014)
Responsible TA: Pearl Pu
Student Name: Renato Kempter
Goals: Design a novel way to recommend interesting places for visiting in the touristic city based on the feedback from social media.
Solution: Web application for exploring trending themes of city locations and finding a set of places to visit for each theme. The locations are extracted from the geo-localized photos of Instagram; and their themes are detected by topics of the associated hashtags.
Keywords: recommender systems, topic modeling, instagram, travel recommendation, system design

People often have a subject, a taste, a style or an interest that guides them to visit places they like. Nevertheless, traditional online travel guides are giving recommendations rather based on the categories of the places and their rankings. In this project, we suggest to use user-generated content to aggregate additional information about locations. Using geo-localized photos from Instagram and their associated hashtags, we develop a system that: a) Automatically clusters a set of instagrams with respect to existent city locations, b) Extracts the trending topics in the hashtags associated with the photos, and c) Generates sets of diverse locations that share the same topic. As topics are extracted directly from the user-generated content, they are more dynamic than predefined category- or tag-based descriptions of locations and reflect user interests and context of visiting those places. The example topics we were able to extract are “Christmas,Christmastree” and “design,vintage,deco”. Finally, we design a web application where users can browse the various topics and their corresponding locations in order to discover locations that suit their taste, style, or interest.

Resources: Demo website

Wine Exploration and Recommendation System


Title: Wine Exploration and Recommendation System
Duration: One semester (September – December 2014)
Responsible TA: Pearl Pu
Student Name: Cédric Rolland
Goals: Design a novel way to recommend and discover wines, which would not require prior knowledge of wine.
Solution: iPhone application for user-centric wine recommendation, exploration and shopping. It features the tailored quiz for taste learning in order to discover wines a user might like. It also includes knowledge quizzes for teaching users wine-related concepts and a reward system for motivating users buying more wines and continuing using the application.
Keywords: recommender systems, interface design, gamification, e-commerce, mobile application design

Wine recommendation is a default feature of multiple websites and applications selling wine. It usually asks the user to input their favorite bottles or give ratings to different wines, and then makes a recommendation based upon this information. However, it can be tedious for users, especially for those with little knowledge about wine. We wanted to design a new way of selling and presenting wine to customers, the way that would not require prior knowledge about wine and would be engaging for users. This resulted in the application called La Caveauté. It recommends wine to users based on the simple quiz about their taste preferences, such as for coffee, breakfast, juices, etc. The user answers to the multiple-choice questions, and the app adds the related tags to the user profile. The algorithm then recommends a wine based on those tags. In addition to the taste quiz and recommending wine, the application provides the full user experience for wine exploration, learning oenology and selecting the most appropriate wine for the moment. The design went through multiple iterations based on the collected user feedback.

Wellness Sensing Using Wearable Sensors


Title: Wellness Sensing Using Wearable Sensors
Duration: One semester (February – June 2012)
Responsible TA: Dr. Zerrin Yumak
Student Name: Javier Martin De Valmaseda
Goals: Develop a stable platform for mobile sensor data collection, labeling and post visualization method.
Solution: Android mobile application for activity labelling and accelerometer data collection.
Keywords: mobile sensors, pervasive healthcare, time series analysis, ubiquitous computing, data collection

Activity recognition of every day activities can be used for many different purposes: preventive medicine, promotion of health-enhancing physical activities and a developing a healthier lifestyle. It has a high industrial interest and a high impact on society. In the same time wearable sensors are becoming less and less intrusive and more accurate. As well, mobile devices are becoming increasingly sophisticated and smartphones with accelerometer and other sensors are widely popular nowadays. In our study data have been collected by 8 users while performing normal daily activities along 2-3 days. They used two different types of sensors: BodyMedia and Affectiva together with an Android phone. We relied on the SAX method to process the data. Afterwards, it can be analyzed by machine learning algorithms to detect certain activity types. SAX is the first method for the symbolic representation of time series that allows dimensionality reduction and indexing with a lower-bounding distancemeasure. This symbolic approach allows a time series of length n to be reduced to a string length w (w < n).


Emotional Kineticons in Online Social Environment


Title: Emotional Kineticons in Online Social Environment
Duration: One semester (February – June 2012)
Responsible TA: Yu Chen
Student Name: Alfredo Cerezo Luna
Goals: Visualize and express emotions in online social environments.
Solution: Embed emotion in Facebook profile pictures.
Keywords: Interface Design, Emotion Visualization
Abstract: We started by defining a library of 9 kineticons according to music related emotions, such as: transcendence, joyful, wonder, tenderness, nostalgia, pacefulness, energy, sadness and tension. We then implemented the kineticons in GroupFun, a group music recommender system that suggests a common playlist for users. Our evaluation was based on a live user study. Aiming at proving that our kineticons are well understood we showed to 15 users our design together with 3 words description of emotion and asked them to vote for the words that best describe each of the 9 kineticons. Additionally, we prepared a video of 9 emotions, each corresponding to a song. All 15 users watched the video and rated (from 1 to 5) the appropriateness of how the kineticons describe the emotions evoked by the song.