Alan RitterPh.d. StudentComputer Science and Engineering University of Washington 
|
|
Overview
I am a 5th year Ph.d. student in the Computer Science Department at the University of Washington.My graduate adviser is Oren Etzioni; I have collaborated with Doug Downey, Mausam, and Stephen Soderland at UW, in addition to Sumit Basu, Colin Cherry and Bill Dolan during internships at Microsoft Research.
Research Summary
Broadly I am interested in getting computers to better understand natural language, and new applications this will enable. I enjoy working with large quantities of text which is not restricted to a narrow domain, such as that found on the Web or in Social Media. Specifically my work focuses on Information Extraction, Computational Lexical Semantics, Latent Variable Models, Language Processing in Social Media and Intelligent Interfaces.Following are short descriptions of specific research directions.
Information Extraction in Social Media
Status Messages written by users of Social Media websites (e.g. Facebook and Twitter) contain a great deal of timely and important information, however there are also many irrelevant and redundant messages which can easily lead to Information Overload. No person can read each of the hundreds of millions of messages produced every day, motivating the need for systems which can automatically extract and aggregate information from these dynamically changing text streams.
Off-the shelf tools such as Part of Speech Taggers and Named Entity Recognizers perform poorly when applied to Social Media text due to it's noisy and unique style. To address this I have been working towards building a set of Twitter-specific text processing tools [EMNLP 2011].
Users of Social Media sites frequently discuss events which will occur in the future. By annotating Named Entities and resolving temporal expressions (for example "next Friday"), we are able to automatically extract a calendar of popular events occurring in the near future from Twitter.Conversational Modeling in Social Media
In addition to discussing upcoming events, users of social networking sites are having public conversations at an unprecedented scale. This presents a unique opportunity to collect millions of naturally occuring conversations and investigate new data-driven techniques for conversational modeling.
I have worked on unsupervised modeling of dialogue acts in Twitter [NAACL 2010]. By remaining agnostic about the set of classes, we are able to learn a model which provides insight into the nature of communication in a new medium.
I have investigated the feasibility of automatically replying to status messages by adapting techniques from Statistical Machine Translation [EMNLP 2011], using millions of naturally occurring Twitter conversations as parallel text. Although there are many differences between conversation and translation, with a few conversation-specific adaptations we are able to build Response Models which often generate appropriate replies to Twitter status posts. This work has several possible applications, including conversationally aware predictive text entry.
A recent talk on this work can be viewed here.
Latent Variable Models of Lexical Semantics
I have applied a variant of Latent Dirichlet Allocation to automatically infer the argument types or Selectional Preferences of textual relations [ACL 2010]. Generative models have the advantage that they provide a principled way to perform many different kinds of probabilistic queries about the data. For example, our model of selectional preferences is useful in filtering improper applications of inference rules in context, showing a substantial improvement over a state-of-the-art rule-filtering system which makes use of a predefined set of classes. The topics discovered by our model can be browsed here. Inference and evaluation code is available for download here.
In addition, I have investigated Distant Supervision with Topic Models. As a distant source of supervision we make use of facts from Freebase, a large, open-domain database, to generate constraints in the topic model. This approach leverages the ambiguous training data provided by Freebase in a principled way, significantly outperforming Co-Training on a weakly supervised named entity classification task [EMNLP 2011].
Utilizing Implicit Feedback in Interactive File Selection
Selection tasks are common in modern computer interfaces; we are often required to select multiple files, emails, and other data entries for copying, modification, deletion etc... Complex selection tasks can require many clicks and mouse movements on behalf of the user; to aid users with these complex selections we propose an interactive machine learning solution [IUI 2009]. In addition to making use of explicit selections and deselections, we utilize implicit features of the user's behaviour such as passing over files, or proximity in the interface. Since the behaviour features are task-independent, we use historical interaction traces as training data. A video demonstration of our file-selection prototype can be viewed here.
Finding Contradictions in Web Text
Many textual relations map one argument to a unique value. For example the verb assassinated should map each direct object to a unique subject. We investigate automatically classifying relation functionality using an unsupervised EM-style algorithm, and evaluate performance at discovering naturally occurring contradictions within a large web corpus [EMNLP 2008]. We show that contradiction detection on the web is a difficult task for a variety of reasons including name ambiguity (e.g. John Smith was born in many different locations), synonyms and meronyms (Mozart was born in both Salzburg and Austria).
Interactive Information Integration with HTML Tables & Freebase
As part of the grad databases class, I investigated applying data-integration techniques to augment HTML tables with additional data from Freebase, in addition to enabling users to quickly verify and contribute data contained in the table. Users can choose to display columns which are not present in the original table, but for which data exists in Freebase, providing direct benefit. They can also easily verify our algorithmically generated mapping from table columns to Freebase attributes, alowing data contained in the table (but missing in Freebase) to be imported. Details on this project including a prototype Firefox browser plugin can be found here.
Teaching Experience
I have served as teaching assistant for the Machine Learning class at UW in Winter 2010 and Winter 2011, where I helped design homework assignments, answered student questions one on one, and presented an hour long lecture in class.
Also at UW, I mentored Sam Clark during his senior year and Master's program; during this time I supervised his work annotating a corpus of Tweets with Parts of Speech,
and building a Part-of-Speech Tagger for Twitter [EMNLP 2011]. Sam is now at Decide.com.
Publications2011
Named Entity Recognition in Tweets: An Experimental Study
Data-Driven Response Generation in Social Media 2010
A Latent Dirichlet Allocation Method for Selectional Preferences
Unsupervised Modeling of Twitter Conversations 2009
Learning to Generalize for Complex Selection Tasks
Filter, Rank, and Transfer the Knowledge: Learning to Chat Alan Ritter, Stephen Soderland, and Oren Etzioni 2009 AAAI Spring Symposium on Learning by Reading and Learning to Read 2008
It's a Contradiction -- No, It's Not: A Case Study using Functional Relations 2006
Distributional Word Clustering in Parallel
Machine Learning Approach to Tuning Distributed Operating System Load Balancing Algorithms
NDIS Network Driver |
Software/Data/Demos |
Presentations
NRL SeminaropenMosix Load Balancing Poster