KIXLAB's research is made possible by generous financial support from KAIST, the National Research Foundation of Korea (NRF), the Institute for Information and Communications Technology Promotion (IITP), LG Electronics, and Samsung Electronics.
Supporting an Iterative Conversation Design Process
Conversation design is an essential step in building a chatbot. Much like designing visual user interfaces, designing conversations benefits from prototyping and testing with users to explore and improve possible and existing conversation flows. However, it can be overwhelming to quickly iterate the conversation design as the iterative process requires not only the conversation design but also building and testing a working chatbot equipped with the conversation. We developed ProtoChat, a prototype system that supports an iterative design of conversation by allowing designers to (1) prototype conversations, (2) test the conversations with crowds, and (3) review and analyze the crowdsourced conversation data. Results of an exploratory study with four conversation designers show that the designers successfully iterated on their conversation design by reviewing how crowds followed the conversation, which provided insights on concrete action items to improve their conversation design.
Snapstream: Snapshot-based Interaction in Live Streaming for Visual Art
Live streaming visual art such as drawing or using design software is gaining popularity. An important aspect of live streams is the direct and real time communication between streamers and viewers. However, currently available text-based interaction limits the expressiveness of viewers as well as streamers, especially when they refer to specific moments or objects in the stream. To investigate the feasibility of using snapshots of streamed content as a way to enhance streamer-viewer interaction, we introduce Snapstream, a system that allows users to take snapshots of the live stream, annotate them, and share the annotated snapshots in the chat. Streamers can also verbally reference a specific snapshot during streaming to respond to viewers’ questions or comments. Results from live deployments show that participants communicate more expressively and clearly with increased engagement using Snapstream. Participants used snapshots to reference part of the artwork, give suggestions on it, make fun images or memes, and log intermediate milestones. Our findings suggest that visual interaction enables richer experiences in live streaming.
Support Listening in Online Space by Reducing Stereotyping
Listening to a wide range of opinions is a virtue for healthy discussion. However, in online space like discussion forum or social media, people can easily filter out perspectives challenging their prior viewpoints. Such filtering is facilitated by people’s stereotypes on people with opposing viewpoints, such as prevalent images on Democrats or Republicans in case of U.S. politics.
We think people can listen to others with different opinions by reducing stereotypes on them. Based on previous work on mitigating stereotypes from social psychology, we are designing an interactive online system called 별별생각, which uses diverse identities of the discussion participants, from demographic ones like age or gender to self-stated identities. With such identities, the user of the system can discover that the opponents have various background beyond their stereotypes. We expect that such intervention can help the users to listen to the people with different opinions more.
Supporting Consensus Building in Collaborative Sequencing
Collaborative Sequencing (Co-Seq) is the process of collaboratively selecting and arranging a set of items into a particular order. It commonly occurs across formal and casual domains (e.g. travel and curriculum planning). As the group must decide on the final sequence, it is desirable to reach consensus. Previous research proposes a consensus building process in which (1) a member makes a proposal, (2) the group evaluates their willingness to commit to it, (3) if they are not, they identify and discuss conflicts, and (4) resolve the conflicts. However, this process can be challenging in Co-Seq due to two main factors: (1) to evaluate a sequence (or proposal), each member must evaluate each of the numerous micro-decisions which constitutes it, and (2) members need to develop and independent understanding of alternatives for each of these micro-decisions. Our goal is to explore avenues through which system support can address these challenges.
Designing to Persuade for Civil Commenting in Online Discussion
To deal with online harassment, e.g., personal attack, stereotyping, profanity, numerous human moderators are working in every discussion site. There are also system support, e.g., automoderator, which are bot account created by human moderators or detection model used by moderators for detecting incivility, insult, profane word in a comment. Moderators can remove the comment, block or ban the abusive user. Although this task of removing, blocking or banning a user can reduce the toxic contents from the discussion, the cost is not low. From the moderators, the task load is too high and the cognitive load is also a major issue. From the user’s perspective, there are lack of transparency and feedback to the user about the moderation process.
With these in mind, the research question we are asking is how can we design interface elements to persuade user in civil commenting and at the same time provide feedback to the user. We are trying to design interface elements to provide real-time feedback to the user, when they are commenting, using machine calculation which outputs are used by the moderators to assist the moderation task. Our expectation is that this real-time feedback will improve the trust and transparency about the moderation process and thus persuade user in civil commenting.
Analyzing K-MOOC Learners' Data for Effective Lecture Design
MOOCs (Massive Open Online Courses) aim to provide a high-quality learning experience to unlimited learners around the world at scale. As lots of learners take courses in MOOC platforms, their learning behaviors are captured in log data that could provide rich insights into how learners collectively learn. Analyzing these logs can benefit (1) learners by providing results from analysis they could self-reflect on, and (2) instructors by enabling them to understand learners and fix the problems in the current instructional material. Despite the benefits of analyzing learners’ behaviors, K-MOOC—a Korean MOOC platform—hasn’t seen much in-depth analysis yet. In this project, we will be analyzing logs from the K-MOOC platform to understand the patterns of lecture video watching pattern and to provide a guideline for an effective lecture design.
Credibility Assessment and Critical Thinking through Microtasks while Reading
Misinformation, which refers to information that is misleading or false, can quickly reach millions of readers with the help of algorithms optimized for engagement and inattentive or malicious users. Although improving the predictive accuracy of ML models is a worthwhile goal, we believe engaging humans into the decision making process should be equally as important. As such, this project explores opportunities in embedding organic crowdsourcing in online discussion domain, as to enable readers in the online platforms to do useful work that benefits both themselves and the platform.
To make this happen, we have two research questions in mind: (1) How to co-optimize the maximization of system-side information gain and user-side engagement gain? and (2) How to motivate readers to engage in more tasks? In order to answer these questions, this project aims to design a crowdsourcing system that recommends a task sequence in order to collect what system needs and interest the readers at the same time.
Predicting Personality via Online and Offline Group-specific Behavior
Personality has a significant influence on individuals’ behavior in co-located groups, affecting group performance, collaboration, and group dynamics. For this reason, many companies and organizations operated by co-located groups utilize personality questionnaires. While self-assessment is widely used to detect personality, it is prone to bias due to self-reporting. Hence, automatic personality assessment (APA), which infers personality by analyzing a person’s behavioral data, is on the rise. However, existing APA systems do not take into account the fact that a person’s behaviors displayed in a group may differ from their behaviors outside the group. Moreover, little research has studied users’ perception towards APA systems: privacy issues and the effect of data collection on their natural behaviors. In order to fill this gap, we present a system that automatically detects the user’s personality in a co-located group by analyzing their group-specific behavioral data both online and offline: user’s online messenger and web/app usage, and offline location and movement. Further, we discuss how the complex relationship between system accuracy, privacy issues, and behavior change should be taken into account when designing APA systems.
SolveDeep: A System for Supporting Subgoal Learning in Online Math Problem Solving
Learner-driven subgoal labeling helps learners form a hierarchical structure of solutions with subgoals, which are conceptual units of procedural problem solving. While learning with such hierarchical structure of a solution in mind is effective in learning problem solving strategies, the development of an interactive feedback system to support subgoal labeling tasks at scale requires significant expert efforts, making learner-driven subgoal labeling difficult to be applied in online learning environments. We propose SolveDeep, a system that provides feedback on learner solutions with peer-generated subgoals. SolveDeep utilizes a learnersourcing workflow to generate the hierarchical representation of possible solutions, and uses a graph-alignment algorithm to generate a solution graph by merging the populated solution structures, which are then used to generate feedback on future learners’ solutions.
Combining Reflection and Practice for Enhancing Creative Learning on Videos
Online videos provide rich materials for people to learn conceptual knowledge, but novices often fail to transfer acquired knowledge into practice. To support learners to apply high-level concepts to the task at hand, this work aims to explore several approaches to enhance learning experience and task performance in the domain of creative tasks (e.g. writing or design). With insights drawn from reflection theories and models, we attempt to integrate reflection and practice into video learning process to enhance learning experience and work performance. Furthermore, we explore the design space of reflective prompting on video learning and the subsequent performance in creative tasks. This research will contributes findings to online video learning and creative tasks.
Many Ideas is a crowd-civic system in which members of a community engage with each other to collect issues, generate ideas, and discuss matters of concern. The system acts as a testbed for researching new design to increase diverse participation. In our first study, we investigated the effect of personalized motivation-supportive messages. Read more
Fall 2017 Undergraduate Theses
Constructing Political Personas that Reflect Voters’ Interests
in Election Promises
Ina Ryu: A Qualitative Analysis of Conversational Techniques in Chatbots
Hyeong Cheol Moon: A Comparison of Korean Sentence Classification Models for Chatbots
Summer 2017 Internship Projects
Learnersourcing Subgoal Labels for Student Solutions
Amy Han: Designing Interactive Distance Cartograms to Support Urban Travelers
Wookjae Byun: User Meets Chatbot for the First Time
Léonore Guillain: RecipeScape: Mining and Analyzing Diverse Processes in Cooking Recipes
Suhwan Jee: Asynchronous-Interactive Feedback Interfaces
Jooyoung Lee: Designing User Interfaces for Supporting Collaborative Decision-Making
Beomsu Kim: Summarizing Image Clusters with Generative Adversarial Nets
Kyungje Jo: Learning Diverse Language Expressions through Video-mining
Hyunwoo Kim: Improving Government Transparency with Social Computing
Jonghyuk Jung: Visualizing Objects and Their Relationships in Living Spaces with Social Media Images
We explore how analyzing cooking recipes in aggregate and in scale helps discovering the core values in the collective cooking culture, and answer fundamental questions like ‘‘what makes a chocolate chip cookie a chocolate chip cookie’’. Aspiring cooks, professional chefs, and cooking hobbyists share their recipes online resulting in thousands of different procedural instructions towards a shared goal. We introduce RecipeScape, a prototype interface which supports visually querying, browsing and comparing cooking recipes at scale. Read more
Rally (아, 쫌!): A Data-driven Community Petition Platform
Often complaints and request from community members’ happen in an ad-hoc manner, in turn, it couldn’t draw the attention from the decision maker. We propose a crowdsourced petition platform Rally to empower the community members’ voice with unified communication pipeline to the decision maker. We focus on the petition which is one of the most common protocols between community members’ and decision makers.
Winter 2017 Intern Projects
Learning Diverse Language Expressions through Video-mining
Jaesung Huh: Assisting essay writing through showing counter-arguments
Deokseong Kim: Constructing personas by using clusters of Instagram user activities
2016 Summer Intern Projects
Improving contents of lecture video by leveraging students’ questions
Hyeungshik Jung: Annotation interface for watching learning video in mobile devices
Dongkwan Kim: Identifying contributing factors of pairwise affinity between politicians
Jiwoo Park: Online Interface that Promotes Higher Level Questions
Taekyung Park: The Effect of Emphasizing Community and Individual Value in Learners’ Engagement in Activities