LKE/KDSL Research Seminar

On Tuesday July 1st 2014, there will be an LKE/KDSL Research Seminar featuring talks by Maryam Tavakol and Jinseok Nam.

Maryam Tavakol:

Title: Factored Markov Decision Processes for Session-Based Recommendation

Abstract: Recommender Systems aim to capture interests of users to provide tailored recommendations. User interests are however often unique and depend on many unobservable factors including a user's mood and the local weather that generates sessions with different intentions. We take a contextual session-based approach and propose a sequential framework using factored Markov decision processes to detect the user's goal of a session. Considering an independence assumption on the attributes of items leads to a set of independent models that can be optimized efficiently using Markov decision processes methods to provide recommendations regarding set of different attributes.

Jinseok Nam:

Title: Exploring label spaces in multi-label learning

Abstract: A textual document is often associated with multiple descriptors, each of which deals with different aspects of the content in the document.

Recently, several work have shown capabilities of neural network based models which are able to capture underlying structures on discrete data such as word embedding.

I will present how such methods can be adapted to cope with multi-label text classification problems, and show the underlying structures over labels.

Afterward you will be introduced to an extension of the methods by placing prior distributions on label spaces in order to explicitly make use of dependencies among labels.