Attention-based Unsupervised Keyphrase Extraction and Phrase Graph for COVID-19 Medical Literature Retrieval

Searching, reading, and finding information from the massive medical text collections are challenging. A typical biomedical search engine is not feasible to navigate each article to find critical information or keyphrases. Moreover, few tools provide a visualization of the relevant phrases to the query. However, there is a need to extract the keyphrases from each document for indexing and efficient search. The transformer-based neural networks—BERT has been used for various natural language processing tasks. The built-in self-attention mechanism can capture the associations between words and phrases in a sentence. This research investigates whether the self-attentions can be utilized to extract keyphrases from a document in an unsupervised manner and identify relevancy between phrases to construct a query relevancy phrase graph to visualize the search corpus phrases on their relevancy and importance. The comparison with six baseline methods shows that the self-attention-based unsupervised keyphrase extraction works well on a medical literature dataset. This unsupervised keyphrase extraction model can also be applied to other text data. The query relevancy graph model is applied to the COVID-19 literature dataset and to demonstrate that the attention-based phrase graph can successfully identify the medical phrases relevant to the query terms.

Understanding Horizon 2020 Data: A Knowledge Graph-Based Approach

This paper aims to meaningfully analyse the Horizon 2020 data existing in the CORDIS repository of EU, and accordingly offer evidence and insights to aid organizations in the formulation of consortia that will prepare and submit winning research proposals to forthcoming calls. The analysis is performed on aggregated data concerning 32,090 funded projects, 34,295 organizations participated in them, and 87,067 public deliverables produced. The modelling of data is performed through a knowledge graph-based approach, aiming to semantically capture existing relationships and reveal hidden information. The main contribution of this work lies in the proper utilization and orchestration of keyphrase extraction and named entity recognition models, together with meaningful graph analytics on top of an efficient graph database. The proposed approach enables users to ask complex questions about the interconnection of various entities related to previously funded research projects. A set of representative queries demonstrating our data representation and analysis approach are given at the end of the paper.

Quantifying Substitutability

<p>In this thesis, we will tackle the problem of how keyphrase extraction systems can be evaluated to reveal their true efficacy. The aim is to develop a new semantically-oriented approximate string matching criteria, one that is comparable to human judgements, but without the cost and energy associated with manual evaluation. This matching criteria can also be adapted for any information retrieval (IR) system where the evaluation process involves comparing candidate strings (produced by the IR system) to a gold standard (created by humans). Our contributions are threefold. First, we define a new semantic relationship called substitutability – how suitable a phrase is when used in place of another – and then design a generic system which measures/quantifies this relationship by exploiting the interlinking structure of external knowledge sources. Second, we develop two concrete substitutability systems based on our generic design: WordSub, which is backed by WordNet; and WikiSub, which is backed by Wikipedia. Third, we construct a dataset, with the help of human volunteers, that isolates the task of measuring substitutability. This dataset is then used to evaluate the performance of our substitutability systems, along with existing approximate string matching techniques, by comparing them using a set of agreement metrics. Our results clearly demonstrate that WordSub and WikiSub comfortably outperform current approaches to approximate string matching, including both lexical-based methods, such as R-precision; and semantically-oriented techniques, such as METEOR. In fact, WikiSub’s performance comes sensibly close to that of an average human volunteer, when comparing it to the optimistic (best-case) interhuman agreement.</p>

Quantifying Substitutability

<p>In this thesis, we will tackle the problem of how keyphrase extraction systems can be evaluated to reveal their true efficacy. The aim is to develop a new semantically-oriented approximate string matching criteria, one that is comparable to human judgements, but without the cost and energy associated with manual evaluation. This matching criteria can also be adapted for any information retrieval (IR) system where the evaluation process involves comparing candidate strings (produced by the IR system) to a gold standard (created by humans). Our contributions are threefold. First, we define a new semantic relationship called substitutability – how suitable a phrase is when used in place of another – and then design a generic system which measures/quantifies this relationship by exploiting the interlinking structure of external knowledge sources. Second, we develop two concrete substitutability systems based on our generic design: WordSub, which is backed by WordNet; and WikiSub, which is backed by Wikipedia. Third, we construct a dataset, with the help of human volunteers, that isolates the task of measuring substitutability. This dataset is then used to evaluate the performance of our substitutability systems, along with existing approximate string matching techniques, by comparing them using a set of agreement metrics. Our results clearly demonstrate that WordSub and WikiSub comfortably outperform current approaches to approximate string matching, including both lexical-based methods, such as R-precision; and semantically-oriented techniques, such as METEOR. In fact, WikiSub’s performance comes sensibly close to that of an average human volunteer, when comparing it to the optimistic (best-case) interhuman agreement.</p>