The Document Vectors Using Cosine Similarity Revisited

TL;DR

This paper revises the evaluation of a simple document vector model, correcting its accuracy, and compares its performance to Transformer models, highlighting its advantages with small datasets and proposing a faster training scheme.

Contribution

The paper identifies an evaluation error in a previous document vector model, corrects its accuracy, and introduces a Bayesian-based sub-sampling scheme to improve training speed and quality.

Findings

Corrected model accuracy from 97.42% to 93.68%.

DV-ngrams-cosine outperforms RoBERTa on small datasets.

Proposed Bayesian sub-sampling accelerates training and enhances performance.

Abstract

The current state-of-the-art test accuracy (97.42\%) on the IMDB movie reviews dataset was reported by \citet{thongtan-phienthrakul-2019-sentiment} and achieved by the logistic regression classifier trained on the Document Vectors using Cosine Similarity (DV-ngrams-cosine) proposed in their paper and the Bag-of-N-grams (BON) vectors scaled by Naive Bayesian weights. While large pre-trained Transformer-based models have shown SOTA results across many datasets and tasks, the aforementioned model has not been surpassed by them, despite being much simpler and pre-trained on the IMDB dataset only. In this paper, we describe an error in the evaluation procedure of this model, which was found when we were trying to analyze its excellent performance on the IMDB dataset. We further show that the previously reported test accuracy of 97.42\% is invalid and should be corrected to 93.68\%. We also analyze the model performance with different amounts of training data (subsets of the IMDB dataset) and compare it to the Transformer-based RoBERTa model. The results show that while RoBERTa has a clear advantage for larger training sets, the DV-ngrams-cosine performs better than RoBERTa when the labelled training set is very small (10 or 20 documents). Finally, we introduce a sub-sampling scheme based on Naive Bayesian weights for the training process of the DV-ngrams-cosine, which leads to faster training and better quality.