Abstract
Topic models and neural networks can discover meaningful low-dimensional latent representations of text corpora; as such, they have become a key technology of document representation. However, such models presume all documents are non-discriminatory, resulting in latent representation dependent upon all other documents and an inability to provide discriminative document representation. To address this problem, we propose a semi-supervised manifold-inspired autoencoder to extract meaningful latent representations of documents, taking the local perspective that the latent representation of nearby documents should be correlative. We first determine the discriminative neighbors set with Euclidean distance in observation spaces. Then, the autoencoder is trained by joint minimization of the Bernoulli cross-entropy error between input and output and the sum of the square error between neighbors of input and output. The results of two widely used corpora show that our method yields at least a 15% improvement in document clustering and a nearly 7% improvement in classification tasks compared to comparative methods. The evidence demonstrates that our method can readily capture more discriminative latent representation of new documents. Moreover, some meaningful combinations of words can be efficiently discovered by activating features that promote the comprehensibility of latent representation.
Highlights
The performance of document analysis and processing systems based on machine learning methods, such as classification[1][2], clustering[3][4], content analysis[5], textual similarity[6], and statistical machine translation (SMT)[7], is heavily dependent on the level of document representation (DR), as different representations may capture and disentangle different degrees of explanatory ingredients hidden in the documents[8]
Supposing that such latent document representation is strongly dependent on its neighbors, from the view of bag of words model, we first represent each document in the forms of a count vector, and select the discriminative neighbors set with Euclidean distance in the observation space
We proposed a semi-supervised manifold-inspired method, namely, the locally embedding autoencoder (LEAE), for document representation
Summary
The performance of document analysis and processing systems based on machine learning methods, such as classification[1][2], clustering[3][4], content analysis[5], textual similarity[6], and statistical machine translation (SMT)[7], is heavily dependent on the level of document representation (DR), as different representations may capture and disentangle different degrees of explanatory ingredients hidden in the documents[8]. Neural networks can capture meaningful latent document representations (i.e., distributed representations) with deep learning techniques, including autoencoders[16], restricted Boltzmann machines (RBMs) [17], neural topic models (NTMs)[18] and document neural autoregressive distribution estimators (DocNADEs)[19] These methods use the word count vector as input and synthesize the input through different hidden layers of various deep neural networks. Topic models and neural networks are embedded with latent factors or topics, preserving the salient statistical structure of intra-documents[19] They represent an improvement for DR, such methods take a global perspective on document space as Euclidean, assuming that all documents are non-discriminatory and indicating that the latent representation is dependent on all other documents. Better representation of the latent document semantics depends on modeling the local document relationship within a neighborhood
Sign-in/Register to view highlights & summary Sign-in/Register to access full text options 
Talk to us
Join us for a 30 min session where you can share your feedback and ask us any queries you have
Schedule a callDisclaimer: All third-party content on this website/platform is and will remain the property of their respective owners and is provided on "as is" basis without any warranties, express or implied. Use of third-party content does not indicate any affiliation, sponsorship with or endorsement by them. Any references to third-party content is to identify the corresponding services and shall be considered fair use under The CopyrightLaw.
