We present a novel approach to learn distributed representation of sentences from unlabeled data by modeling both content and context of a sentence. The content model learns sentence representation by predicting its words. On the other hand, the context model comprises a neighbor prediction component and a regularizer to model distributional and proximity hypotheses, respectively. We propose an online algorithm to train the model components jointly. We evaluate the models in a setup, where contextual information is available. The experimental results on tasks involving classification, clustering, and ranking of sentences show that our model outperforms the best existing models by a wide margin across multiple datasets.

@inproceedings{saha-joty-hasan-ecml-17, abstract = {We present a novel approach to learn distributed representation of sentences from unlabeled data by modeling both content and context of a sentence. The content model learns sentence representation by predicting its words. On the other hand, the context model comprises a neighbor prediction component and a regularizer to model distributional and proximity hypotheses, respectively. We propose an online algorithm to train the model components jointly. We evaluate the models in a setup, where contextual information is available. The experimental results on tasks involving classification, clustering, and ranking of sentences show that our model outperforms the best existing models by a wide margin across multiple datasets.}, address = {Macedonia, Skopje}, author = {Tanay Saha and Shafiq Joty and Mohammad Hasan}, booktitle = {Proceedings of The European Conference on Machine Learning & Principles and Practice of knowledge discovery in databases}, month = {September}, pages = {xx--xx}, publisher = {Springer}, series = {ECML-PKDD'17}, title = {CON-S2V: A Generic Framework for Incorporating Extra-Sentential Context into Sen2Vec}, url = {papers/saha-joty-hasan-ecml-17.pdf}, year = {2017} }