Unsupervised Graph-Based Similarity Learning Using Heterogeneous Features.

摘要

Relational data refers to data that contains explicit relations among objects. Nowadays, relationaldata are universal and have a broad appeal in many different application domains. Theproblem of estimating similarity between objects is a core requirement for many standardMachine Learning (ML), Natural Language Processing (NLP) and Information Retrieval(IR) problems such as clustering, classiffication, word sense disambiguation, etc. Traditionalmachine learning approaches represent the data using simple, concise representations suchas feature vectors. While this works very well for homogeneous data, i.e, data with a singlefeature type such as text, it does not exploit the availability of dfferent feature types fully.For example, scientic publications have text, citations, authorship information, venue information.Each of the features can be used for estimating similarity. Representing suchobjects has been a key issue in efficient mining (Getoor and Taskar, 2007). In this thesis,we propose natural representations for relational data using multiple, connected layers ofgraphs; one for each feature type. Also, we propose novel algorithms for estimating similarityusing multiple heterogeneous features. Also, we present novel algorithms for tasks like topic detection and music recommendation using the estimated similarity measure. Wedemonstrate superior performance of the proposed algorithms (root mean squared error of24.81 on the Yahoo! KDD Music recommendation data set and classiffication accuracy of88% on the ACL Anthology Network data set) over many of the state of the art algorithms,such as Latent Semantic Analysis (LSA), Multiple Kernel Learning (MKL) and spectralclustering and baselines on large, standard data sets.