Text Analytics: Semi-supervised Learning for Semantic Text Processing
Description
Machine Learning methods that are able to learn from both labeled and unlabeled data
are commonly subsumed under the term semi-supervised learning. Semi-supervised learning is a very important approach for many practical applications, also in text processing, like relation extraction and sentiment analysis.
- Typically, only few labeled data are available for training, while unlabeled data are abundant.
- Often, labeled data is given only for a particular domain, while for the domain of interest there is a lack in labeled data.
The seminar reviews the most important methods for semi-supervised learning in the field of semantic text processing. Topics include bootstrapping methods, self-learning, distant supervision, semi-supervised methods in Deep Learning, and their usage in information extraction tasks (e.g. relation extraction) and sentiment analysis.
Expectation
Each student is expected to
- attend the seminar sessions and actively contribute to the discussion in the seminar
- prepare a presentation on a topic relevant for the seminar
- present this presentation and be able to answer questions from the audience
- prepare a term paper on the topic
Organization
- Seminar sessions are on Thursdays, 13:30-15:10 in S105/22
- All other information will be provided during the course and added here.
The first seminar will be on October 16.
Literature
- XiaojinZhu and Andrew B.Goldberg. Introduction to Semi-Supervised Learning. Synthesis Lectures on Artificial Intelligence and Machine Learning 2009 3:1. Morgan & Claypool.
- Anders Søgaard. Semi-Supervised Learning and Domain Adaptation in Natural Language Processing. Synthesis Lectures on Human Language Technologies 2013 6:2. Morgan & Claypool.
Timetable
Introductory lectures on Natural Language Processing and Semi-supervised Learning will be held in the first three regular sessions of the seminar on Thursdays (16.10., 23.10. and 6.11.) 13:30-15:10 in S105/22.
On October 30, there will be a guest lecture by Prof. Jordan Boyd-Graber, title: Thinking on your Feet: Reinforcement Learning for Incremental Language Tasks
The program for the remainder of the seminar will be announced according to number of participants and topics to be discussed.
Lecturers
- Dr. Judith Eckle-Kohler (office hours: will be announced in the first session, please resigster by e-mail)
- Silvana Hartmann
- Prof. Dr. Iryna Gurevych