We investigate in user behavior by mapping them to the tasks they are trying to accomplish on a daily basis.


Towards Task Understanding in Visual Settings
Sebastin Santy, Wazeer Zulfikar, Rishabh Mehrotra, Emine Yilmaz
AAAI Conference on Artificial Intelligence, 2019

Understanding tasks being undertaken in a scene. Leveraging insights from real world task understanding systems, to propose a framework composed of convolutional neural networks, and an external hierarchical task ontology to produce task descriptions from input images.

Leveraging Task Information for Question Recommendation
Sebastin Santy, Rishabh Mehrotra, Emine Yilmaz
Under Review

Leverage an existing hierarchical task ontology in order to understand user behaviour from the task perspective. A lot of daily searches performed involve a user being interested in accomplishing a particular task. Hence, leveraging task information will help in question recommendation.

Task Hierarchy 138K
Sebastin Santy, Rishabh Mehrotra, Emine Yilmaz

A hierarchy containing tasks segregated according to the category they fall in. This hierarchy describes the task-subtask relationship and how there exists several levels of tasks. These task artifacts are collected from Wikihow articles which can be effectively interpreted as a task in order to build a task hierarchy.

Dynamic Enrichment of Existing Ontology in an online manner
Sebastin Santy, Rishabh Mehrotra, Emine Yilmaz

Several websites maintain a large hierarchy of categories in order to segregate their pages and provide useful information to the user. These websites range from answer websites like Quora to e-commerce websites like Amazon. However, addition of new categories to this existing hierarchy is often manual. We investigate in updation of hierarchy by using the ground-truth hierarchy and learning the pre-decided paths for making future decisions.

Task Images 1.5M
Sebastin Santy, Rishabh Mehrotra, Emine Yilmaz

Each article in Wikihow contains images belonging to each task being solved in it. These images are a great source in order to understand the tasks being undertaken in scenes. This dataset is a collection of 1.5 million of such images annotated with their corresponding task.

Aspect-based Characterization of Community-based Question Answering Platforms
Sebastin Santy, Prasanta Bhattacharya, Rishabh Mehrotra, Emine Yilmaz
Ongoing work

We depend on several different platforms to accomplish our tasks. We subconciously select the platforms which might help in solving a particular task. Search Engines adds a layer of carelessness for task-relevance of platforms. We put effort in categorizing the platforms based on what task they focus on solving the most.

Characterizing online task heterogeneity on language-localized Q&A Platforms
Sebastin Santy, Prasanta Bhattacharya, Rishabh Mehrotra, Emine Yilmaz
Ongoing work

Several online articles are manually translated to different languages. Effort is put into translating only if it is relevant in a particular linguistic community. We analyze on what tasks are relevant in those communities by observing these patterns.