IMPROVING ALGORITHMS, FOSTERING COMMUNITY
| Photos by Evan Krape and courtesy of Women in Electrical and Computer Engineering | February 13, 2023 | UDaily Article Link
With innovative research and inclusivity, Prof. Hui Fang exemplifies UD’s collaboration culture
Whether it’s looking up directions to a nearby restaurant or checking the score of the basketball game, search engines are a helpful tool for navigating everyday life. But for certain tasks, datasets or platforms, different algorithms from the ones that power internet search engines are needed.
Pioneering fundamental research in this field is Hui Fang, a professor in the University of Delaware’s Department of Electrical and Computer Engineering. Her group is working on new ways to help researchers find what they need in their data while developing approaches to ensure that search results are fair and unbiased. Fang is also a champion of community building in UD’s College of Engineering and exemplifies what it means to be part of UD’s “collaboration culture.”
Making sense of massive amounts of information
Fang’s research is focused on information retrieval, the process of how people access, manage, and use information. Through their fundamental research on this topic, Fang and her team are focused on ways to “help users access information from various data sources,” she explained.
“The main motivation for my research is the problem of information overload, where there are so many different types and amounts of information, and how we help the user deal with that overload,” Fang said.
While search engines like Google are powerful tools for scouring the internet, users’ information needs often go far beyond finding relevant web pages, explained Fang.
“Finding relevant information is usually the first step. Many information-seeking tasks, such as writing a literature survey about an emerging research topic, require users to synthesize relevant information from multiple documents and analyze the content for knowledge discovery or decision making,” she said. “Because one strategy won’t work for all information tasks, in our lab we are studying how to develop technologies to help users combat the information overload problems encountered in different disciplines.”
To this end, Fang’s research group regularly works with collaborators in other disciplines, creating customized search platforms and information management tools. Her previous efforts include projects on using smartphones to monitor clinical depression, analyzing political behavior on mobile devices, and an app to address opioid use disorders.
Building custom search platforms for their collaborators also allows her group to incorporate state-of-the-art search algorithms. One such method that her group specializes in is axiomatic thinking, a method that provides a theoretical framework for optimizing search engines in a way that is guided by a set of retrieval constraints.
“As different disciplines’ information tasks vary significantly in many aspects — such as data sources, relevance criteria and granularity of the search results, etc. — a customized information system is necessary to ensure the results are satisfying,” Fang said. “Axiomatic thinking provides guidance on how the customization should be done through formulating domain knowledge as retrieval constraints.”
Enhancing disaster research
One example of how Fang’s research is used to create customized search platforms is her collaboration with UD’s Disaster Research Center (DRC), where her team developed a new information retrieval system tailored to the DRC’s unique datasets and research needs. In particular, the system allows researchers to more easily manage both digital and physical data, and also links together different data types, such as news articles and tweets, so that information is easier to parse through during an ongoing disaster.
“This project has resulted in an increase in the ability of disaster researchers to access material from our collections, and put scholars and scholarship about disasters in conversation with each other that would otherwise have remained siloed,” said Valerie Marlowe, assistant director of archives and collections in the DRC. “We regularly hear from researchers that they are organically finding material that they would never have thought to search for in the past, which is a huge benefit to an interdisciplinary field seeking to address the challenges that will result from increasing disasters in the face of climate change.”
Kuang Lu, who is a key developer of the system, earned his doctorate in 2019 and is now a research scientist at Meta.
“With this project, I created a tool that people use and that’s helping the DRC do their research in a more effective, efficient way, and that’s the most rewarding part of this work,” Kuang Lu said.
Mining plastics
Fang’s research can also be applied for problems related to data mining, a process where users find interesting patterns in their data without needing a specific search query. “The user may not have a clear understanding of what they are looking for, but they want to discover an interesting pattern or gain knowledge from their collection,” Fang said.
To this end, Fang is working with Prof. Dionisios Vlachos, both of whom are part of the Center for Plastics Innovation, to mine the chemical scientific literature. Their goal is to identify existing research gaps and potentially discover new ways to upcycle plastics.
Yue Zhang, a doctoral student working with Fang, is developing a platform to help researchers more easily search through existing journal articles to find chemicals of interest and reaction parameters for further study. He is also working with Cong Wang, a postdoc in Vlachos’ group, on annotating a dataset to help build a “chemicals search engine” for catalysts, which speed up chemical reactions and are a key component of process optimization.
“Information on chemical transformations is complex — it includes text, tables, and graphs; extracting information from all these formats is daunting,” Vlachos said. “The representation of chemicals and solid catalysts is non standardized, making teaching a search engine difficult. Being able to search thousands of publications and patents through the work Hui and her students develop could significantly accelerate the discovery of transformations to convert plastics waste, a major societal and environmental threat, to valuable products.”
Zhang enjoys the interdisciplinary nature of this project, where he gets to work at the forefront of information extraction while also learning about chemical engineering, as well as the opportunity to develop a new tool. “Dr. Fang gives me a lot of freedom and encourages me to try new things,” Zhang said about what it’s like to work in her lab.
Avoiding bias, ensuring fairness
Fang is also interested in ensuring that search results are unbiased, relevant, and provide a balanced perspective of the viable information that’s available. But while the potential benefits of algorithmic fairness are wide-reaching, from amplifying underrepresented groups to correcting historical biases, defining fairness in this context is complicated.
To come up with a better assessment of fairness, Fumian Chen, a doctoral student in the Financial Services Analytics (FSAN) program, is working on a Wikipedia search project that aims to provide a more balanced exposure to different groups and cover more perspectives within search results.
“Estimating fairness is very challenging, and even choosing what fairness measures you use to evaluate your algorithms is also a challenge,” he said. “In this project, we try to bring more flexibility and interpretability to this fairness using advanced machine learning methods.”
Fang said she hopes to expand on this and other related projects in the future and is especially interested in the topic of information filtering and how personalized recommendation algorithms can lead to unintentional echo chambers. “In the education domain, this means that kids can essentially live in information bubbles without even realizing,” she said. “I’m interested in developing information systems that can counteract these negative effects.”
Building community for Women in ECE
Along with a research program that focuses on applying state-of-the-art search methods to real-world problems, Fang is also a champion of diversity, equity and inclusion efforts in ECE.
“A conversation I had with one of our freshman made me realize that the pandemic made everything isolated and that we should start a student group that brings all the women students together in this field,” said Fang about the impetus to help put together a group of students who could lead Women in Electrical and Computer Engineering (WECE).
Now, WECE is an active student organization that has been hosting social and professional events since the spring of 2021. Recurring events include study breaks, game nights, resume workshops and Women in Industry panels.
“We started with what we wanted to achieve with the group, and because we were remote it forced us to focus on our goal and our mission, which was to build a community,” said Susan Arnopolin, a 2022 graduate who served as the WECE co-president in 2021-2022 and is now working at the Johns Hopkins Applied Physics Laboratory near Baltimore. “Dr. Fang was really crucial in the process, because she really brought us all together.”
“We stepped out as a fully formed group in a way that a lot of clubs or organizations really don’t get a chance to do,” said Ellie Rosin, former WECE co-president who also graduated in 2022 and is now working at Northrop Grumman. “Dr. Fang was pivotal to the planning, she had ideas and enthusiasm, and key was confidence that people would want to be there.”
“Professor Fang has been a great mentor and has positively impacted many WECE members’ career development,” added co-president of WECE Nafisa Maryam. “She has also been a strong influence in promoting WECE’s participation in professional development programs and conferences. Her support and guidance to WECE’s growth has been immeasurable and speaks for itself through the incredible success WECE has achieved so far.”
Being an example of UD’s ‘collaboration culture’
“Professor Fang is a leading scientist in the area of information retrieval and data sciences with a true gift for using these powerful techniques to address major challenges facing society such as climate change and sustainability,” said Jamie Phillips, Chair of the ECE department. “Hui is a shining example for developing the collaborative community that we aim to foster in ECE and at UD. More broadly, she has had a transformative impact on mentoring and building an enthusiastic cohort of women engineers through our WECE student group.”
Fang said she was initially attracted to UD and its ECE department because of its collegial environment and “collaboration culture,” which is an essential component of her group’s research that regularly reaches across disciplines and fields.
“The most challenging part of the work is to understand the terminology and language of another discipline,” she said. “But, thankfully, I have great collaborators who work with us to understand new concepts, and great students who work with me to learn new disciplines.”