Uma Sawant
uma AT cse DOT iitb DOT ac DOT in
Infolab, Kresit lab 212,
IIT Bombay, Powai,
Mumbai 400 076
http://www.cse.iitb.ac.in/~uma/
I am currently a PhD candidate in IIT Bombay, with Prof. Soumen Chakrabarti and Prof. Ganesh Ramakrishnan as my principal advisers.
Research Topic
Web search views the Web as a collection of documents and returns a list of documents ranked in the order of relevance
in response to a user query. Identification of entities, i.e., objects or concepts in queries and documents (e.g. people,
places, cars) allows us to do more; such as providing direct answers to many user queries. I am interested in this kind of
"Entity-aware search". I am currently studying entity annotation, query interpretation and ranking components for the same.
Experience
- Research intern at Yahoo! labs, barcelona (Sept 2014 - Nov 2014)
I worked on improving entity search results by considering the interaction between entity annotator and ranker
components. My hosts were Dr. Peter Mika and Dr. Roi Blanco.
- Research engineer at Yahoo! labs, bangalore (July 2008 - July 2011)
I worked on Yahoo! search projects which involved understanding user intent and feedback, converting data to
features and building ranking models. I have worked on Yahoo! Web search and image search ranking models, which
have been applied in
production.
Publications
- Mandar Joshi, Uma Sawant, Soumen Chakrabarti, Knowledge Graph and Corpus Driven Segmentation and Answer Inference for
Telegraphic Entity-seeking Queries. In Conference on Empirical Methods in Natural Language Processing (EMNLP) 2014, Doha, Quatar.
- Uma Sawant, Soumen Chakrabarti, Learning Joint Query Interpretation and Response Ranking.
In proceedings of
22nd international World Wide Web Conference (WWW 2013), Rio De Janeiro, Brazil.
- Uma Sawant, Soumen Chakrabarti, Features and Aggregators for Web-scale Entity Search. Technical report, 2013.
- Soumen Chakrabarti, Rajiv Khanna, Uma Sawant, Chiru Bhattacharyya, Structured Learning for Non-Smooth
Ranking Losses. In proceedings of the 14th ACM SIGKDD International Conference on Knowledge Discovery and
Data Mining (KDD-2008), Las Vegas.
- Aniket Dalal, Kumar N., Uma Sawant, Sandeep Shelke and Pushpak Bhattacharyya, Building Feature Rich POS
Tagger for Morphologically Rich Languages: Experiences in Hindi. In proceedings of 5th International
Conference on Natural Language Processing (ICON 2007), IIIT, Hyderabad.
Talks
- An overview of query interpretation and ranking for entity-aware search. Seminar at IIT Kanpur, August 2014.
- Joint query interpretation and response ranking for entity-aware search. Presentation at Yahoo summer school, June 2013.
Education
- Ongoing Ph.D. in Computer Science from IIT Bombay, Powai with CPI 8.64/10.0 (2011 - to date)
- M. Tech. in Computer Science from IIT Bombay, Powai with CPI 9.71/10.0 (2008).
- B. E. in Information Technology from Sardar Patel College of Engineering, Mumbai with 69% (2005).
- H.S.C. from Sathaye College, Mumbai with 93.50% (2001).
- S.S.C. from Paranjape Vidyalaya, Mumbai with 88.93% (1999).
Achievements
- Yahoo Superstar Award (Team) to Image Search team, 2010.
- Recipient of the Google India Women in Engineering Award, 2008.
- Part of the 4-member team of IIT Bombay which secured second prize in NLPAI-ML 2006 contest for shallow
parsing of Indian languages, a national level contest in the field of natural language processing.
- Secured 99.63 percentile with All India Rank 96 in the Graduate Aptitude Test in Engineering (the entrance
examination for the IITs), Test of Computer Science (2005).
- Ranked second in all streams of the Sardar Patel College of Engineering in first year of engineering (2001-2002).
- Ranked 15th in the merit list by Maharashtra Board in H.S.C. (2001).
- Recipient of the National Talent Search Scholarship awarded by the Government of India (1999).
Previous Research Projects
- Learning to rank for non-smooth ranking losses (Masters thesis, under the guidance of Prof. Soumen Chakrabarti, IITB)
Our objective was to learn a real-valued ranking function from the given labeled training data. Using structured
learning paradigm, we directly optimize for non-smooth ranking losses like Mean Reciprocal Rank(MRR) and Normalized
Discounted Cumulative Gain
(NDCG) etc. The challenge here is to avoid the complexity in training
structured learner
by intelligently choosing training instances.
- Supervised learning of web-pages using visual layout as features, (with Prof. Soumen Chakrabarti, IITB)
Our objective was to extract the names and values of entity attributes
from product web pages. A web page was
interpreted as an undirected graph
with various visual and structural properties. We created an extractor,
by defining
an undirected graphical model on the web page.
Selected Course Projects and Seminars
- Entity Ranking and Evidence Aggregation. (with Prof. Soumen Chakrabarti, IITB)
In this seminar, I studied various challenges involved in Entity Ranking over
unstructured corpus. The supporting
evidence for a potential answer entity can
come from different documents across the corpus. I considered multiple
methods
to evaluate an evidence snippet based on signals like term-entity proximity and term rarity and finally
aggregate the evidence to induce
ranking over entities.
- Part-of-speech tagger for Hindi, (with Prof. Pushpak Bhattacharyya, IITB)
We built a statistical part-of-speech (POS) tagger for Hindi based on the Maximum Entropy Markov Model. The
tagger used language independent as well as language specific features.
- Sort order optimization using functional dependencies in the Volcano optimizer, (with Prof. S. Sudarshan, IITB)
A query containing "order by" or "group by" clauses imposes
a sort order on the output given by the attributes in
these clauses.
We extended the Volcano optimizer to exploit the functional dependencies in
reducing sort orders. In
turn, this enabled the optimizer to choose the
less expensive plans corresponding to the smaller sort orders.
- Statistical natural language modeling, (with Prof. Soumen Chakrabarti, IITB)
A statistical natural language model captures the statistical view of the
natural language generation process. In this
seminar, I studied in detail
various popular natural language models, followed by an overview of the
discriminative
approach to language modeling.
Courses
Algorithms and complexity, Statistical foundations of machine learning, Web search
and mining, Data mining, Graphical
models, Convex Optimization, Natural language
processing, Implementation techniques in relational databases, Advanced
databases
Technical Skills
- Programming Skills: Java, C, C++, SQL, Matlab, Scilab, Shell scripting
- Database Systems: PGSQL, Oracle, Mysql
Extra-Curricular Activities
- Technical Secretary for hostel 11, IITB, in year 2007-2008.
- Member of the organizing committee of SPACE, the cultural
event and Nirmaan, the technical event of Sardar Patel
College of Engineering in 2003.
- Interests include music, books, swimming.