Abstract

Nanopublications are RDF graphs encoding scientific facts extracted from the literature and enriched with provenance and attribution information. There are millions of nanopublications currently available on the Web, especially in the life science domain. Nanopublications are thought to facilitate the discovery, exploration, and re-use of scientific facts. Nevertheless, they are still not widely used by scientists outside specific circles; they are hard to find and rarely cited. We believe this is due to the lack of services to seek, find, and understand nanopublications' content. To this end, we present the NanoWeb application to seamlessly search, access, explore, and re-use the nanopublications publicly available on the Web. For the time being, NanoWeb focuses on the life science domain where the vastest amount of nanopublications are available. It is a unified access point to the world of nanopublications enabling search over graph data, direct connections to evidence papers, and scientific curated databases, and visual and intuitive exploration of the relation network created by the encoded scientific facts.

Abstract

Search Engines (SE) have been shown to perpetuate well-known gender stereotypes identified in psychology literature and to in uence users accordingly. Similar biases were found encoded in Word Embeddings (WEs) learned from large online corpora. In this context, we propose the Gender Stereotype Reinforcement (GSR) measure, which quantifies the tendency of a SE to support gender stereotypes, leveraging gender-related information encoded in WEs. Through the critical lens of construct validity, we validate the proposed measure on synthetic and real collections. Subsequently, we use GSR to compare widely-used Information Retrieval ranking algorithms, including lexical, semantic, and neural models. We check if and how ranking algorithms based on WEs inherit the biases of the underlying embeddings. We also consider the most common debiasing approaches for WEs proposed in the literature and test their impact in terms of GSR and common performance measures. To the best of our knowledge, GSR is the first specifically tailored measure for IR, capable of quantifying representational harms.

Abstract

It is widely accepted that data is fundamental for research and should therefore be cited as textual scientific publications. However, issues like data citation, handling and counting the credit generated by such citations, remain open research questions. Data credit is a new measure of value built on top of data citation, which enables us to annotate data with a value, representing its importance. Data credit can be considered as a new tool that, together with traditional citations, helps to recognize the value of data and its creators in a world that is ever more depending on data.

In this paper we define Data Credit Distribution (DCD) as a process by which credit generated by citations is given to the single elements of a database. We focus on a scenario where a paper cites data from a database obtained by issuing a query. The citation generates credit which is then divided among the database entities responsible for generating the query output. One key aspect of our work is to credit not only the explicitly cited entities, but even those that contribute to their existence, but which are not accounted in the query output.

We propose a data Credit Distribution Strategy (CDS) based on data provenance and implement a system that uses the information provided by data citations to distribute the credit in a relational database accordingly. As use case and for evaluation purposes, we adopt the IUPHAR/BPS Guide to Pharmacology (GtoPdb), a curated relational database. We show how credit can be used to highlight areas of the database that are frequently used. Moreover, we also underline how credit rewards data and authors based on their research impact, and not merely on the number of citations. This can lead to designing new bibliometrics for data citations.

Abstract

The semantic mismatch between query and document terms – i.e., the semantic gap – is a long-standing problem in Information Retrieval (IR). Two main linguistic features related to the semantic gap that can be exploited to improve retrieval are synonymy and polysemy. Recent works integrate knowledge from curated external resources into the learning process of neural language models to reduce the effect of the semantic gap. However, these knowledge-enhanced language models have been used in IR mostly for re-ranking and not directly for document retrieval.

We propose the Semantic-Aware Neural Framework for IR (SAFIR), an unsupervised knowledge-enhanced neural framework explicitly tailored for IR. SAFIR jointly learns word, concept, and document representations from scratch. The learned representations encode both polysemy and synonymy to address the semantic gap. SAFIR can be employed in any domain where external knowledge resources are available. We investigate its application in the medical domain where the semantic gap is prominent and there are many specialized and manually curated knowledge resources. The evaluation on shared test collections for medical literature retrieval shows the effectiveness of SAFIR in terms of retrieving and ranking relevant documents most affected by the semantic gap.

Abstract

Keyword-based access to structured data has been gaining traction both in research and industry as a means to facilitate access to information. In recent years, the research community and big data technology vendors have put much effort into developing new approaches for keyword search over structured data. Accessing these data through structured query languages, such as SQL or SPARQL, can be hard for endusers accustomed to Web-based search systems. To overcome this issue, keyword search in databases is becoming the technology of choice, although its efficiency and effectiveness problems still prevent a large scale diffusion. In this work, we focus on graph data, and we propose the TSA+BM25 and the TSA+VDP keyword search systems over RDF datasets based on the “virtual documents” approach. This approach enables high scalability because it moves most of the computational complexity off-line and then exploits highly efficient text retrieval techniques and data structures to carry out the on-line phase. Nevertheless, text retrieval techniques scale well to large datasets but need to be adapted to the complexity of structured data. The new approaches we propose are more efficient and effective compared to state-of-the-art systems. In particular, we show that our systems scale to work with RDF datasets composed of hundreds of millions of triples and obtain competitive results in terms of effectiveness.

Abstract

This paper analyzes two state-of-the-art Neural Information Retrieval (NeuIR) models: the Deep Relevance Matching Model (DRMM) and the Neural Vector Space Model (NVSM).

Our contributions include: (i) a reproducibility study of two state-of-the-art supervised and unsupervised NeuIR models, where we present the issues we encountered during their reproducibility; (ii) a performance comparison with other lexical, semantic and state-of-the-art models, showing that traditional lexical models are still highly competitive with DRMM and NVSM; (iii) an application of DRMM and NVSM on collections from heterogeneous search domains and in different languages, which helped us to analyze the cases where DRMM and NVSM can be recommended; (iv) an evaluation of the impact of varying word embedding models on DRMM, showing how relevance-based representations generally outperform semantic-based ones; (v) a topic-by-topic evaluation of the selected NeuIR approaches, comparing their performance to the well-known BM25 lexical model, where we perform an in-depth analysis of the different cases where DRMM and NVSM outperform the BM25 model or fail to do so.

We run an extensive experimental evaluation to check if the improvements of NeuIR models, if any, over the selected baselines are statistically significant.