scite_
Research Supported Answers for Search
With a team of 3, I designed scite_’s new “Ask A Question” feature to allow both academics and non-academics alike the ability to receive research-backed answers to their direct question searches
OVERVIEW
scite_ is a platform for discovering and evaluating scientific articles to see how a publication has been cited and give the context of the citation by describing whether it provides supporting or contrasting evidence for the claim. For this specific project, we were tasked with designing a new feature on scite_, the “ask a question” (AAQ) feature. This is a feature where users can ask direct questions to Scite’s search feature and receive answers backed by research.
Over the course of the project, our team conducted business and user research to determine how both the existing audience of scite’s platform as well as new audiences might use the AAQ feature. We determined three primary locations on the website that the AAQ feature would need to be designed into: the home page, search results page, and the browser extension. We then designed and tested multiple iterations of these pages until we landed on our final design to present to scite’s team.
MY ROLE
DURATION
Product Designer
4 Weeks
THE TEAM
Ry Dunn - UX Designer
Krystal Riccardi - UX Designer Bowei Wang - UX Designer
TOOLS
CLIENT TEAM
Figma
Figjam
Otter
Zoom
Zeplin
scite_ CEO
scite_ Product Designer
scite_ Engineering Team
The CHALLENGE
The challenge of our project was to redesign scite_’s new AAQ feature and determine its positioning within the website architecture to increase usage, improve the legibility of its results, and align the feature with scite_’s future vision of expanding its audience to non-academic casual users. We determined our challenge statement to be:
"How might we design an intuitive interface that translates complex research findings into an easily understandable format for all audiences?"
THE SOLUTION
A clearer way to search for research backed answers to direct questions.
We worked to deliver a solution that met the needs of both existing users and prospective users, with features including:
1
Creating a pathway to AAQ directly from the home page
2
3
Updating iconography and simplifying search results to improve clarity for those without research experience
Implementing the AAQ feature directly into the browser extension to allow users to search from external pages
4
Splitting the result card layout to separate metadata from the result and make the context badge more prominent
APPROACH TO DESIGN
RESEARCH
Understand the user the business, the market, and the technology deeply
SYNTHESIZE
Strip away all of the excess to reveal the goals, needs, and challenges that exist
IDEATE
Design only for what is necessary and useful; then make it undeniably beautiful
DELIVER
Document, organize, and deliver for a smooth developer handoff
RESEARCH & SYNTHESIZE
MARKET RESEARCH
Research of the client's business as well as its competitors and comparators was conducted to understand the landscape of the product
USABILITY TESTING
Usability testing was run on the existing website to determine, both quantitatively and qualitatively, the usability and intuitiveness of the product
HEURISTIC EVALUATION
A heuristic evaluation using the Abby Method was conducted to evaluate the current site based on ten standard heuristics
USER INTERVIEWS
In-depth user interviews were conducted to learn directly from the target audiences about their goals, needs and challenges within the problem space
MARKET RESEARCH
Before diving into scite’s new AAQ feature, we wanted to gain a deep understanding of the natural language processing (NLP) models that drive direct question search-and-answer and the way that scite’s algorithm interacts with a user’s input to determine a response. Our team developed a visual representation of the search journey using NLP and did market and competitor research into existing NLP interfaces (GPT, Stable Diffusion, Dall-E, Grammarly, etc.) to develop a better understanding of the general technological landscape that scite’s AAQ feature was operating within.
Key Takeaways
NLP Inner Workings
Natural language processing models use various techniques (e.g. parsing, NER, stop word removal, etc.) to determine the intent behind a search question input and produce meaningful results for the user
Clarity & Legibility
Because we did not have the capacity to rework scite_’s underlying answer formulation algorithm, we would focus our efforts on providing the most clear and legible results with the search results produced by scite_’s existing NLP model
Unique Value
scite_’s unique value lies not in the ability to answer a direct question but, rather its ability to support or refute that claim with scholarly literature surrounding a given topic area through its smart citations
HEURISTIC EVALUATION
Our team conducted a heuristic evaluation on the existing site as a way to take a detailed look at the individual elements of the home and search pages primarily. Through this method, we determined that the areas with the largest room for improvement were those shown below in red.
USABILITY TESTING
How did the existing website perform with users in usability testing?
To gain a deeper insight into the way users moved through the existing website and where AAQ might operate most effectively, we decided to conduct a round of usability testing with the product as it stood. We had users complete 7 total tasks and asked for qualitative feedback throughout as well to understand their choices and preferences. Below are the key takeaways we learned during our testing, broken down by screen.
Home Page
Search bar is not visually prominent enough
Cannot access AAQ from home page
Explainer GIF is confusing and doesn't push action
Search Results Page
Search filters and categories are overwhelming
Citation badge is too small and icons lack clarity
Not enough distinction between title line and results
Extension Explainer Page
Excessive use of text makes the page overwhelming
Lack of visuals creates a significant delay of action
Primary functions of the extension are not laid out
TESTING INSIGHTS TO FEATURES
People engaged in casual research tend to be miffed by academic language, and may seek alternative resources in more plain language
People want to be able to easily access additional information from a scholarly paper to make a judgement on its credibility
People often begin their search with a keyword search of a broad topic area and then develop their research question
Clean up the search results cards for “Ask a Question” so that the answer is clearer and stands out from additional card information
Add the ability to expand the abstract of a paper directly in the search card results without the need to exit scite_ to an external page
Add functionality to the browser extension to be able to ask a question directly through scite_ while on external websites
USER INTERVIEWS
GOAL OF RESEARCH
The goal of our research was primarily to explore when in the research process the Ask a Question feature is used most often, if there is a use for the feature for a broader audience, how academics and non-academics alike interact with research and what information is necessary for users asking direct questions to receive? To determine the answers to these questions, we interviewed 2 primary audiences; one being professional academics or researchers and the other being non-academic professionals whose research is casual in nature.
CHALLENGE
“How might users who don’t fit into scite’s existing demographic benefit from receiving research-backed answers to direct question searches?”
WHAT DID THE USERS SAY?
KEY FINDINGS
EASY VERIFICATION
People who engage in casual research desire a convenient way to access credible sources that can be easily verified and cross-referenced.
CLEAR LANGUAGE
When searching for the answer to a question, people tend to want the answer presented in clear language without necessarily needing to open the article.
QUESTION TIMING
People engaged in research often begin their process with a broad question and narrow down the scope as they move forward with their search.
MULTIPLE SOURCES
When searching for the answer to a question, people like to have multiple sources.
WHO ARE WE DESIGNING FOR?
After conducting user interviews and synthesizing our research findings into key takeaways, we put together two fictitious user personas for our two primary audiences to serve as a guidepost as we entered the ideation and design phase.
WHERE CAN WE HELP OUR PERSONAS?
With our personas in mind, we developed two journey maps to understand the emotional curve as our users go through the research process. This helped our team identify the points in the process where there was room for improvement.
IDEATE
ASK A QUESTION FEATURE
Scite_’s existing value proposition is in its powerful natural language processing (NLP) algorithm, which provides its users with a platform to expand their research and display credibility by providing context from surrounding academic literature. Our team set out to define the positioning of AAQ within scite_’s broader value. Through our user research, we determined that this tool would continue to serve academics as another tool in their research process but, that its primary value was to an audience with less experience with scientific research. With this in mind, we had to continue to present the academic research first, while making the overall usability of the feature simple enough that non-researchers could access the vale as well.
LOW FIDELITY DESIGNS
Prior to building out our high fidelity designs in Figma, our team spent time ideating in a formal design studio. During this time we sketched ideas for the home page, results pages, and browser extension to dump as many ideas on the table as we could before we ultimately pared them down to our final choices.
HIGH FIDELITY WIREFRAMES
With our final design choices made, we began to build them in high fidelity to be able to effectively test them with users. Since several of the changes we were making were dependent on color and visual design, we decided to move straight into high fidelity instead of making mid-fidelity wireframes.
HOME PAGE
ASK A QUESTION RESULTS
SEARCH ALL RESULTS
BROWSER EXTENSION EXPLAINER PAGE
BROWSER EXTENSION WIDGET
USABILITY TESTING
After our usability testing on the existing website, we ran two additional rounds of testing on our proposed designs using a high-fidelity prototype. The results below reflect the deltas between the existing website and the second round of usability testing.
Home Page
Search bar brought to center of page
AAQ suggested pathway added directly to search bar
Dynamic CTA carousel added to encourage search
Search Results Page
Metadata split from result for visual clarity
Citation badge explainer added for clarity of function
Expandable abstract keeps users on scite_ page
Extension Explainer Page
Visuals added for quick understanding of functions
Reduced amount of text for visual breathing room
Three primary functions are clearly displayed
DELIVER
IMPROVEMENTS & FINAL DESIGNS
Below are detailed explanations of the improvements made between the existing site and our final high fidelity designs based on our user research and usability testing findings.
Users’ attention is not directed towards existing search bar
Users are unable to discern the purpose of the excerpt and are overwhelmed by the amount of text on the results cards
Users do not pay attention to citation badge and associate “?” icon with help and are unsure of the “mentioning” icon
Users don’t read browser extension page because it is too text-heavy
Users struggle to find widget because it blends in with the background of academic websites
Make search bar more prominent through repositioning and color
Split the result card layout to separate metadata from result and made research context badge more prominent.
Change the “?” icon for contrasting citations to “!” and the mentioning icon for the quote bubble
Reduce text on browser extension page and visualize its functionality
Implement color and adjust sizing to make widget stand out on academic websites
HIGH FIDELITY PROTOTYPE
OUTCOMES + NEXT STEPS
LIMITATIONS
As with any project, there were some limitations that our group faced in as we moved through the design process. Those primary limitations can be found below.
With additional time, we would have loved the ability to interview users from other demographics to determine if our findings were consistent across different potential audiences. Furthermore, another round of usability testing on the most current card designs could have been done to affirm our final design decisions.
Our secondary limitation was in our task creation for our usability testing. Without our design solution, we developed the tasks for the existing site. However, we found that we needed to add additional tasks to the usability tests of our redesign and therefore did not have concrete data to compare from.
NEXT STEPS
1
Continue with another round of usability testing on the most current prototype.
2
Apply the findings from the heuristic evaluation that we conducted across the site to improve learnability and clarity
3
Conduct tree testing on the navigational hierarchy to improve the information architecture.
WHAT DID I LEARN?
Throughout this project, I found that striking the balance of designing for the current state of the NLP algorithm’s capabilities and scite_’s user base while still maintaining the long term vision of the team to expand the marketability of scite_ to non-academic audiences to be extremely stimulating. While challenging at times, it was an exciting puzzle to produce a usable design that was accessible to all audiences while still providing the depth of information for academics who needed it. Furthermore, in this project I developed a far deeper appreciation for what I call “technology landscape research”. The tech behind AI, and in particular NLP algorithms is certainly not simple. However, by spending enough time to grasp the mechanics of how things like text parsing, stop word removal, and named entry recognition work at a high level, I was able to create designs that pushed the boundaries of the technology without being completely unrealistic of what would ultimately be possible within the scope we were working within. I found this step at the beginning of our project immensely useful and continued to reference it as we moved through each phase. Overall, it was a privilege to work on this project and be a part of a company who is bringing important academic literature to people in a digestible format.