Bobble Concept Storyboard

Bobble is a chrome extension that detects malicious attacks known as pollution attacks in personalized search results. Bobble detects these attacks by verifying your result set for a search query against multiple profiles across the globe. 

Problem Statement 
An attacker can pollute the input to the personalization engine present in Google search and thereby control the output of content presented, as a result manipulate user behavior. How might we detect these polluted search results so that people can make well-informed decisions free from manipulation? 

As a design research assistant, I was advised by Dr. Marshini Chetty, HCI professor at University of Maryland and we collaborated with Dr. Xinyu Xing, CS professor at Georgia Institute of Technology and his assistant Dr. Mshab Alrizah

Research Goals 
Within our research project of designing user-facing tools that detect pollution attacks in personalized search results, we had two subgoals: 
a. Understand how people perceive personalization in content-based systems such as Google search. 
b. Investigate whether people can detect manipulated search results.

Study Design Summary 
To realistically simulate the manipulation, we conducted a month-long user study with 20 participants which consisted of one online survey, two in-person interviews, data log collection for 24 search queries that we had given to our participants. Furthermore, all participants installed a version of Bobble based on their group i.e. control or experiment group. Based on the randomly assigned group participants’ search results were/were not manipulated. We conducted our analysis using quantitative and qualitative results from user interviews and data logs. 

Significant Results 
Our research work in addition to designing the application produced three key significant results namely:
a. Incomplete understanding of search personalization - Results demonstrated the skewed understanding of personalized information systems, currently heavily restricted to product based recommendation engines and barely extending towards content-based services such as search results, news feed etc. The lack of awareness and incomplete understanding of content based personalized information services such as search results provides significant opportunity to administer a pollution attack.
b. Detection of polluted results requires a user-facing tool - Post our analysis we were able to triangulate that participants in the experiment group were unable to detect manipulated search results. This coupled with other results supported our hypothesis that there is a need for user-facing tools that detect polluted results in search queries.

Prototype - Round One 
Our design work was intentional and was based on the research findings. For the prototype, after our exploration, we designed our persona using our participant data. We presented our sketches to participants and received feedback.
Prototype - Round Two 
After incorporating suggestions from our participants we decided to use maps for our medium fidelity prototype given its ease of understanding and significance. A light-yellow highlighted search return is one of the top 10 search return that appears on others' web browsers but not on the user's results; A light-green highlighted search return is one of the top 3 search returns that appears on others' web browsers but not on the user's browser.

Round Two Prototype

Round Two Workflow Diagram

Our work presented at the annual CHI 2016 Conference


Back to Top