The quality of word embeddings depends on the input corpora, model architectures, and hyper-parameter settings. Using the state-of-the-art neural embedding tool word2vec and both intrinsic and extrinsic evaluations, we present a comprehensive study of how the quality of embeddings changes according to these features. Apart from identifying the most influential hyper-parameters, we also observe one that creates contradictory results between intrinsic and extrinsic evaluations. Furthermore, we find that bigger corpora do not necessarily produce better biomedical domain word embeddings. We make our evaluation tools and resources as well as the created state-of-the-art word embeddings available under open licenses from https://github.com/ cambridgeltl/BioNLP-2016.

BackgroundNamed Entity Recognition (NER) is a key task in biomedical text mining. Accurate NER systems require task-specific, manually-annotated datasets, which are expensive to develop and thus limited in size. Since such datasets contain related but different information, an interesting question is whether it might be possible to use them together to improve NER performance. To investigate this, we develop supervised, multi-task, convolutional neural network models and apply them to a large number of varied existing biomedical named entity datasets. Additionally, we investigated the effect of dataset size on performance in both single- and multi-task settings.ResultsWe present a single-task model for NER, a Multi-output multi-task model and a Dependent multi-task model. We apply the three models to 15 biomedical datasets containing multiple named entities including Anatomy, Chemical, Disease, Gene/Protein and Species. Each dataset represent a task. The results from the single-task model and the multi-task models are then compared for evidence of benefits from Multi-task Learning.With the Multi-output multi-task model we observed an average F-score improvement of 0.8% when compared to the single-task model from an average baseline of 78.4%. Although there was a significant drop in performance on one dataset, performance improves significantly for five datasets by up to 6.3%. For the Dependent multi-task model we observed an average improvement of 0.4% when compared to the single-task model. There were no significant drops in performance on any dataset, and performance improves significantly for six datasets by up to 1.1%.The dataset size experiments found that as dataset size decreased, the multi-output model’s performance increased compared to the single-task model’s. Using 50, 25 and 10% of the training data resulted in an average drop of approximately 3.4, 8 and 16.7% respectively for the single-task model but approximately 0.2, 3.0 and 9.8% for the multi-task model.ConclusionsOur results show that, on average, the multi-task models produced better NER results than the single-task models trained on a single NER dataset. We also found that Multi-task Learning is beneficial for small datasets. Across the various settings the improvements are significant, demonstrating the benefit of Multi-task Learning for this task.Electronic supplementary materialThe online version of this article (doi:10.1186/s12859-017-1776-8) contains supplementary material, which is available to authorized users.

The quality of word representations is frequently assessed using correlation with human judgements of word similarity. Here, we question whether such intrinsic evaluation can predict the merits of the representations for downstream tasks. We study the correlation between results on ten word similarity benchmarks and tagger performance on three standard sequence labeling tasks using a variety of word vectors induced from an unannotated corpus of 3.8 billion words, and demonstrate that most intrinsic evaluations are poor predictors of downstream performance. We argue that this issue can be traced in part to a failure to distinguish specific similarity from relatedness in intrinsic evaluation datasets. We make our evaluation tools openly available to facilitate further study.

No abstract

BackgroundWord representations support a variety of Natural Language Processing (NLP) tasks. The quality of these representations is typically assessed by comparing the distances in the induced vector spaces against human similarity judgements. Whereas comprehensive evaluation resources have recently been developed for the general domain, similar resources for biomedicine currently suffer from the lack of coverage, both in terms of word types included and with respect to the semantic distinctions. Notably, verbs have been excluded, although they are essential for the interpretation of biomedical language. Further, current resources do not discern between semantic similarity and semantic relatedness, although this has been proven as an important predictor of the usefulness of word representations and their performance in downstream applications.ResultsWe present two novel comprehensive resources targeting the evaluation of word representations in biomedicine. These resources, Bio-SimVerb and Bio-SimLex, address the previously mentioned problems, and can be used for evaluations of verb and noun representations respectively. In our experiments, we have computed the Pearson’s correlation between performances on intrinsic and extrinsic tasks using twelve popular state-of-the-art representation models (e.g. word2vec models). The intrinsic–extrinsic correlations using our datasets are notably higher than with previous intrinsic evaluation benchmarks such as UMNSRS and MayoSRS. In addition, when evaluating representation models for their abilities to capture verb and noun semantics individually, we show a considerable variation between performances across all models.ConclusionBio-SimVerb and Bio-SimLex enable intrinsic evaluation of word representations. This evaluation can serve as a predictor of performance on various downstream tasks in the biomedical domain. The results on Bio-SimVerb and Bio-SimLex using standard word representation models highlight the importance of developing dedicated evaluation resources for NLP in biomedicine for particular word classes (e.g. verbs). These are needed to identify the most accurate methods for learning class-specific representations. Bio-SimVerb and Bio-SimLex are publicly available.

No abstract

BackgroundVerbNet, an extensive computational verb lexicon for English, has proved useful for supporting a wide range of Natural Language Processing tasks requiring information about the behaviour and meaning of verbs. Biomedical text processing and mining could benefit from a similar resource. We take the first step towards the development of BioVerbNet: A VerbNet specifically aimed at describing verbs in the area of biomedicine. Because VerbNet-style classification is extremely time consuming, we start from a small manual classification of biomedical verbs and apply a state-of-the-art neural representation model, specifically developed for class-based optimization, to expand the classification with new verbs, using all the PubMed abstracts and the full articles in the PubMed Central Open Access subset as data.ResultsDirect evaluation of the resulting classification against BioSimVerb (verb similarity judgement data in biomedicine) shows promising results when representation learning is performed using verb class-based contexts. Human validation by linguists and biologists reveals that the automatically expanded classification is highly accurate. Including novel, valid member verbs and classes, our method can be used to facilitate cost-effective development of BioVerbNet.ConclusionThis work constitutes the first effort on applying a state-of-the-art architecture for neural representation learning to biomedical verb classification. While we discuss future optimization of the method, our promising results suggest that the automatic classification released with this article can be used to readily support application tasks in biomedicine.Electronic supplementary materialThe online version of this article (10.1186/s13326-018-0193-x) contains supplementary material, which is available to authorized users.

Word representations are mathematical objects that capture the semantic and syntactic properties of words in a way that is interpretable by machines. Recently, encoding word properties into low‐dimensional vector spaces using neural networks has become increasingly popular. Word embeddings are now used as the main input to natural language processing (NLP) applications, achieving cutting‐edge results. Nevertheless, most word‐embedding studies are carried out with general‐domain text and evaluation datasets, and their results do not necessarily apply to text from other domains (e.g., biomedicine) that are linguistically distinct from general English. To achieve maximum benefit when using word embeddings for biomedical NLP tasks, they need to be induced and evaluated using in‐domain resources. Thus, it is essential to create a detailed review of biomedical embeddings that can be used as a reference for researchers to train in‐domain models. In this paper, we review biomedical word embedding studies from three key aspects: the corpora, models and evaluation methods. We first describe the characteristics of various biomedical corpora, and then compare popular embedding models. After that, we discuss different evaluation methods for biomedical embeddings. For each aspect, we summarize the various challenges discussed in the literature. Finally, we conclude the paper by proposing future directions that will help advance research into biomedical embeddings.