An Empirical Evaluation of doc2vec with Practical Insights into Document Embedding Generation
This work addresses reproducibility challenges for researchers and practitioners using document embeddings, offering practical insights and tools.
The paper tackled the reproducibility issues of doc2vec by conducting an empirical evaluation on two tasks, finding that it performs robustly with large external corpora and pre-trained word embeddings, and providing hyper-parameter recommendations and released source code.
Recently, Le and Mikolov (2014) proposed doc2vec as an extension to word2vec (Mikolov et al., 2013a) to learn document-level embeddings. Despite promising results in the original paper, others have struggled to reproduce those results. This paper presents a rigorous empirical evaluation of doc2vec over two tasks. We compare doc2vec to two baselines and two state-of-the-art document embedding methodologies. We found that doc2vec performs robustly when using models trained on large external corpora, and can be further improved by using pre-trained word embeddings. We also provide recommendations on hyper-parameter settings for general purpose applications, and release source code to induce document embeddings using our trained doc2vec models.