Nav Nudge: An Interaction Technique for Visually Reducing the Feature Search Space on Demand

Based on the findings from previous studies, we started designing an interaction technique that reduces the search space of a feature-rich UI using visual cues, on demand to nudge older adults in the right direction.

In this step, we implemented a fully functioning prototype and conducted a usability test with older users.

Here is a summary of the process up to this point.

1. Identify the common challenges that older adults face when interacting with mobile interfaces. DISCOVER HERE

2. Capture design requirements based on observed challenges and user behaviors, then review them through the design probe. DISCOVER HERE

3. Explore effective visual presentation methods for older adults, and design the application. DISCOVER HERE

4. Develop a prototype app for Android devices, and validate the solution through user testing.

Feature Search Space Reduction Technique

To reduce the number of features that older adults have to search to find one, we created a feature search space reduction technique using natural-language processing (NLP) methods.

Our feature search space reduction technique consists of three steps:

1. Determining what function the user is looking for from a verbal query.
2. Extracting features available in the user’s current UI view, along with their location.
3. Identifying what UI features match the most with the user’s query.

Implementation

• Nav Nudge was designed on the Android 13 platform.
• we used OpenAI’s text-davinci-002 model (Chat GPT 3.0), by directly sending prompts to the model from within the app
  (“Extract a keyword from the following sentence: [INPUT-PHRASE], Keyword:”).
• We used the Universal Sentence Encoder (universal-sentence-encoder v4) to determine a semantic similarity score between the two sets of keywords (user’s query and UI elements).
• A Dedicated server was established that communicated with the app via a REST API.

Task-focused exploratory study &
Semi-structured interview

To determine whether this Nav Nudge would be utilized by older users when they encountered challenges during UI exploration, we conducted a small task-focused study to explore how older adults would use Nav Nudge outside the lab, in an unconstrained setting.

To evaluate Nav Nudge, we chose a mobile app that would be conceptually familiar to older adults but not commonly used — Organic Maps.
It allowed us to focus on app navigation challenges rather than problems stemming from unfamiliarity with the concept of the app.

Tasks

1. Find out how long it will take to travel from Chicago O’Hare Airport to Chicago Union Station using public transit.
2. Identify nearby hotels to Chicago Union Station without using a search box and bookmark one of them.
3. Share your current location with [first author’s email address].
4. Remove the hotel that you bookmarked in Task 2 from your saved list.

Participants

• 10 participants.
• 60 years old or above living in Metro-Chicago area.
• All participants were familiar with map apps, but no experience with Organic Maps.

Settings

• A place of participant’s choice.
• Seated with no restrictions on their posture.
• Participants used provided Android phones. Device settings were customized for each participant.
• No time limit, but encouraged to complete it as quickly as possible.

Data Analysis

• Atlas.ti was used for video coding.
• Two experts independently identified instances of interaction events.
• Used Cohen's kappa to quantify the level of agreement (Cohen’s κ = .98, p < .001).

95% of users queries were answered properly by Nav Nudge.

• 8 out of 10 participants asked at least one question, and 19 questions were recorded.
• In 18 out of 19 instances, the desired feature was included in the reduced space.
• One instance of failure occurred when a participant hesitated during a question (“Where can I send...”),
  leading to a voice recognizer timeout (1.5 seconds).
• Here are three examples of cases where the nav nudge was used.

Older adults used Nav Nudge when they were getting lost.

• 80% of participants used the voice assistant.
• 75% of getting lost were followed by the question (15 out of 20 getting losts).