Search

Knowledge base question answering (KBQA)is an important task in Natural Language Processing. Existing approaches face significant challenges including complex question understanding, necessity for reasoning, and lack of large end-to-end training datasets. In this work, we propose Neuro-Symbolic Question Answering (NSQA), a modular KBQA system, that leverages (1) Abstract Meaning Representation (AMR) parses for task-independent question understanding; (2) a simple yet effective graph transformation approach to convert AMR parses into candidate logical queries that are aligned to the KB; (3) a pipeline-based approach which integrates multiple, reusable modules that are trained specifically for their individual tasks (semantic parser, entity andrelationship linkers, and neuro-symbolic reasoner) and do not require end-to-end training data. NSQA achieves state-of-the-art performance on two prominent KBQA datasets based on DBpedia (QALD-9 and LC-QuAD1.0). Furthermore, our analysis emphasizes that AMR is a powerful tool for KBQA systems., Accepted to Findings of ACL

Drug–Drug Interactions (DDIs) are a major cause of preventable Adverse Drug Reactions (ADRs), causing a significant burden on the patients’ health and the healthcare system. It is widely known that clinical studies cannot sufficiently and accurately identify DDIs for new drugs before they are made available on the market. In addition, existing public and proprietary sources of DDI information are known to be incomplete and/or inaccurate and so not reliable. As a result, there is an emerging body of research on in-silico prediction of drug–druginteractions. In this paper, we present Tiresias , a large-scale similarity-based framework that predicts DDIs through link prediction. Tiresias takes in various sources of drug-related data and knowledge as inputs, and provides DDI predictions as outputs. The process starts with semantic integration of the input data that results in a knowledge graph describing drug attributes and relationships with various related entities such as enzymes, chemical structures, and pathways. The knowledge graph is then used to compute several similarity measures between all the drugs in a scalable and distributed framework. In particular, Tiresias utilizes two classes of features in a knowledge graph: local and global features. Local features are derived from the information directly associated to each drug (i.e., one hop away) while global features are learnt by minimizing a global loss function that considers the complete structure of the knowledge graph. The resulting similarity metrics are used to build features for a large-scale logistic regression model to predict potential DDIs. We highlight the novelty of our proposed Tiresias and perform thorough evaluation of the quality of the predictions. The results show the effectiveness of Tiresias in both predicting new interactions among existing drugs as well as newly developed drugs.

Textual entailment is a fundamental task in natural language processing. Most approaches for solving the problem use only the textual content present in training data. A few approaches have shown that information from external knowledge sources like knowledge graphs (KGs) can add value, in addition to the textual content, by providing background knowledge that may be critical for a task. However, the proposed models do not fully exploit the information in the usually large and noisy KGs, and it is not clear how it can be effectively encoded to be useful for entailment. We present an approach that complements text-based entailment models with information from KGs by (1) using Personalized PageR- ank to generate contextual subgraphs with reduced noise and (2) encoding these subgraphs using graph convolutional networks to capture KG structure. Our technique extends the capability of text models exploiting structural and semantic information found in KGs. We evaluate our approach on multiple textual entailment datasets and show that the use of external knowledge helps improve prediction accuracy. This is particularly evident in the challenging BreakingNLI dataset, where we see an absolute improvement of 5-20% over multiple text-based entailment models.

Background Adverse drug reactions (ADRs) are unintended and harmful reactions caused by normal uses of drugs. Predicting and preventing ADRs in the early stage of the drug development pipeline can help to enhance drug safety and reduce financial costs. Methods In this paper, we developed machine learning models including a deep learning framework which can simultaneously predict ADRs and identify the molecular substructures associated with those ADRs without defining the substructures a-priori. Results We evaluated the performance of our model with ten different state-of-the-art fingerprint models and found that neural fingerprints from the deep learning model outperformed all other methods in predicting ADRs. Via feature analysis on drug structures, we identified important molecular substructures that are associated with specific ADRs and assessed their associations via statistical analysis. Conclusions The deep learning model with feature analysis, substructure identification, and statistical assessment provides a promising solution for identifying risky components within molecular structures and can potentially help to improve drug safety evaluation.

Natural Language Inference (NLI) is fundamental to many Natural Language Processing (NLP) applications including semantic search and question answering. The NLI problem has gained significant attention thanks to the release of large scale, challenging datasets. Present approaches to the problem largely focus on learning-based methods that use only textual information in order to classify whether a given premise entails, contradicts, or is neutral with respect to a given hypothesis. Surprisingly, the use of methods based on structured knowledge -- a central topic in artificial intelligence -- has not received much attention vis-a-vis the NLI problem. While there are many open knowledge bases that contain various types of reasoning information, their use for NLI has not been well explored. To address this, we present a combination of techniques that harness knowledge graphs to improve performance on the NLI problem in the science questions domain. We present the results of applying our techniques on text, graph, and text-to-graph based models, and discuss implications for the use of external knowledge in solving the NLI problem. Our model achieves the new state-of-the-art performance on the NLI problem over the SciTail science questions dataset., 9 pages, 3 figures, 5 tables

While much work has focused on efficient processing of Big Data, little work considers how to understand them. In this paper, we describe Helix, a system for guided exploration of Big Data. Helix provides a unified view of sources, ranging from spreadsheets and XML files with no schema, all the way to RDF graphs and relational data with well-defined schemas. Helix users explore these heterogeneous data sources through a combination of keyword searches and navigation of linked web pages that include information about the schemas, as well as data and semantic links within and across sources. At a technical level, the paper describes the research challenges involved in developing Helix, along with a set of real-world usage scenarios and the lessons learned.

Linked Data on the web consists of over 1000 datasets from a variety of domains. They are queried with the SPARQL query language. There exist many implementations of SPARQL, and this rich ecosystem has demanded a precise specification and compliance tests. However, the SPARQL specification has grown in complexity, and it is increasingly difficult for developers to validate their implementations. In this paper, we present a declarative specification for SPARQL, based on relational logic. It describes SPARQL with just a few operators, and is executable: queries written in it can be directly executed against real datasets.

Drug-Drug Interactions DDIs are a major cause of preventable adverse drug reactions ADRs, causing a significant burden on the patients' health and the healthcare system. It is widely known that clinical studies cannot sufficiently and accurately identify DDIs for new drugs before they are made available on the market. In addition, existing public and proprietary sources of DDI information are known to be incomplete and/or inaccurate and so not reliable. As a result, there is an emerging body of research on in-silico prediction of drug-drug interactions. We present Tiresias, a framework that takes in various sources of drug-related data and knowledge as inputs, and provides DDI predictions as outputs. The process starts with semantic integration of the input data that results in a knowledge graph describing drug attributes and relationships with various related entities such as enzymes, chemical structures, and pathways. The knowledge graph is then used to compute several similarity measures between all the drugs in a scalable and distributed framework. The resulting similarity metrics are used to build features for a large-scale logistic regression model to predict potential DDIs. We highlight the novelty of our proposed approach and perform thorough evaluation of the quality of the predictions. The results show the effectiveness of Tiresias in both predicting new interactions among existing drugs and among newly developed and existing drugs.

SPARQL has many nice features for accessing data integrated across different data sources, which is an important step in any data analysis task. We report on the use of SPARQL for two real data analytic use cases from the healthcare and life sciences domains, which exposed certain weaknesses in the current specification of SPARQL, specifically when the data being integrated is most conveniently accessed via RESTful services and in formats beyond RDF, such as XML. We therefore extended SPARQL with generalized service, constructs for accessing services beyond the SPARQL endpoints supported by service. For efficiency, our constructs support posting data, which is also not supported by service. We provide an open source implementation of this SPARQL endpoint in an RDF store called Quetzal, and evaluate its use in the two data analytic scenarios over real datasets.

We show that existing mature, relational optimizers can be exploited with a novel schema to give better performance for property graph storage and retrieval than popular noSQL graph stores. The schema combines relational storage for adjacency information with JSON storage for vertex and edge attributes. We demonstrate that this particular schema design has benefits compared to a purely relational or purely JSON solution. The query translation mechanism translates Gremlin queries with no side effects into SQL queries so that one can leverage relational query optimizers. We also conduct an empirical evaluation of our schema design and query translation mechanism with two existing popular property graph stores. We show that our system is 2-8 times better on query performance, and 10-30 times better in throughput on 4.3 billion edge graphs compared to existing stores.

A major challenge in designing neural network (NN) systems is to determine the best structure and parameters for the network given the data for the machine learning problem at hand. Examples of parameters are the number of layers and nodes, the learning rates, and the dropout rates. Typically, these parameters are chosen based on heuristic rules and are manually fine-tuned, which may be very time-consuming, because evaluating the performance of a single parametrization of the NN may require several hours. In this paper, we address the problem of choosing appropriate parameters for the NN by formulating it as a box-constrained mathematical optimization problem and applying a derivative-free optimization tool that automatically and effectively searches the parameter space. The optimization tool employs a radial basis function model of the objective function (the prediction accuracy of the NN) to accelerate the discovery of configurations yielding high accuracy. Candidate configurations explored by the algorithm are trained to a small number of epochs, and only the most promising candidates receive full training. The performance of the proposed methodology is assessed on benchmark sets and in the context of predicting drug–drug interactions, showing promising results. The optimization tool used in this paper is open source.

In graph databases, a given graph query can be executed in a large variety of semantically equivalent ways. Each such execution plan produces the same results, but at different computation costs. The query planning problem consists of finding, for a given query, an execution plan with the minimum cost. The traditional greedy or heuristic cost-based approaches addressing the query planning problem do not guarantee by design the optimality of the chosen execution plan. In this paper, we present a principled framework to solve the query planning problem by casting it into an Integer Linear Programming problem, and discuss its applications to testing and improving heuristic-based query planners.

Although RDF graphs have schema information associated with them, in practice it is very common to find cases in which data do not fully conform to their schema. A prominent example of this is DBpedia, which is RDF data extracted from Wikipedia, a publicly editable source of information. In such situations, it becomes interesting to study the structural properties of the actual data, because the schema gives an incomplete description of the organization of a dataset. In this paper we have approached the study of the structuredness of an RDF graph in a principled way: we propose a framework for specifying structuredness functions, which gauge the degree to which an RDF graph conforms to a schema. In particular, we first define a formal language for specifying structuredness functions with expressions we call rules. This language allows a user or a database administrator to state a rule to which an RDF graph may fully or partially conform. Then we consider the issue of discovering a refinement of a sort (type) by partitioning the dataset into subsets whose structuredness is over a specified threshold. In particular, we prove that the natural decision problem associated to this refinement problem is NP-complete, and we provide a natural translation of this problem into Integer Linear Programming (ILP). Finally, we test this ILP solution with two real world datasets, DBpedia Persons and WordNet Nouns, and 4 different and intuitive rules, which gauge the structuredness in different ways. The rules give meaningful refinements of the datasets, showing that our language can be a powerful tool for understanding the structure of RDF data., 18 pages, 8 figures. To be published in PVLDB Vol. 8, No. 9

In coalition operations, information from different sources belong to different organisations have to be gathered and aggregated. The information from these resources may not be consistent. Inconsistencies in the gathered information creates severe uncertainties that hinders the usefulness of the information. In this paper, we have propose a Subjective Logic based approach for modelling the trustworthiness of information sources within a specific context. This model is used to handle inconsistencies through filtering information from less trustworthy sources.

In this paper, we describe a mechanism for ontology alignment using instance based matching of types (or classes). Instance-based matching is known to be a useful technique for matching ontologies that have different names and different structures. A key problem in instance matching of types, however, is scaling the matching algorithm to (a) handle types with a large number of instances, and (b) efficiently match a large number of type pairs. We propose the use of state-of-the art locality-sensitive hashing (LSH) techniques to vastly improve the scalability of instance matching across multiple types. We show the feasibility of our approach with DBpedia and Freebase, two different type systems with hundreds and thousands of types, respectively. We describe how these techniques can be used to estimate containment or equivalence relations between two type systems, and we compare two different LSH techniques for computing instance similarity.

The size, heterogeneity and dynamicity of data within an enterprise makes indexing, integration and analysis of the data increasingly difficult tasks. On the other hand, there has been a massive increase in the amount of high-quality open data available on the Web that could provide invaluable insights to data analysts and business intelligence specialists within the enterprise. The goal of Helix project is to provide users within the enterprise with a platform that allows them to perform online analysis of almost any type and amount of internal data using the power of external knowledge bases available on the Web. Such a platform requires a novel, data-format agnostic indexing mechanism, and light-weight data linking techniques that could link semantically related records across internal and external data sources of various characteristics. We present the initial architecture of our system and discuss several research challenges involved in building such a system.

Decision makers (humans or software agents alike) are faced with the challenge of examining large volumes of information originating from heterogeneous sources with the goal of ascertaining trust in various pieces of information. In this paper we argue (using examples) that traditional trust models are limited in their data model by assuming a pair-wise numeric rating between two entities (e.g., eBay recommendations, Netflix movie rating, etc). We present a novel trust computational model for rich, complex and uncertain information encoded using Bayesian Description Logics. We present security and scalability tradeoffs that arise in the new model, and the results of an evaluation of the first prototype implementation under a variety attack scenarios.

A key problem in ontology alignment is that different ontological features (e.g., lexical, structural or semantic) vary widely in their importance for different ontology comparisons. In this paper, we present a set of principled techniques that exploit user feedback to customize the alignment process for a given pair of ontologies. Specifically, we propose an iterative supervised-learning approach to (i) determine the weights assigned to each alignment strategy and use these weights to combine them for matching ontology entities; and (ii) determine the degree to which the information from such matches should be propagated to their neighbors along different relationships for collective matching. We demonstrate the utility of these techniques with standard benchmark datasets and large, real-world ontologies, showing improvements in F-scores of up to 70% from the weighting mechanism and up to 40% from collective matching, compared to an unweighted linear combination of matching strategies without information propagation.

A common vocabulary is vital to smooth business operation, yet codifying and maintaining an enterprise vocabulary is an arduous, manual task. We describe a process to automatically extract a domain specific vocabulary (terms and types) from unstructured data in the enterprise guided by term definitions in Linked Open Data (LOD). We validate our techniques by applying them to the IT (Information Technology) domain, taking 58 Gartner analyst reports and using two specific LOD sources --- DBpedia and Freebase. We show initial findings that address the generalizability of these techniques for vocabulary extraction in new domains, such as the energy industry.

In this paper, we describe scalable highly expressive reasoner (SHER), a breakthrough technology that provides semantic querying of large relational datasets using OWL ontologies. SHER relies on a unique algorithm based on ontology summarization and combines a traditional in-memory description logic reasoner with a database backed RDF Store to scale reasoning to very large Aboxes. In our latest experiments, SHER is able to do sound and complete conjunctive query answering up to 7 million triples in seconds, and scales to datasets with 60 million triples, responding to queries in minutes. We describe the SHER system architecture, discuss the underlying components and their functionality, and briefly highlight two concrete use-cases of scalable OWL reasoning based on SHER in the Health Care and Life Science space. The SHER system, with the source code, is available for download (free for academic use) at: http://www.alphaworks.ibm.com/tech/sher.

Conjunctive query answering for $\mathcal{EL}^{++}$ ontologies has recently drawn much attention, as the Description Logic $\mathcal{EL}^{++}$ captures the expressivity of many large ontologies in the biomedical domain and is the foundation for the OWL 2 EL profile. In this paper, we propose a practical approach for conjunctive query answering in a fragment of $\mathcal{EL}^{++}$, namely acyclic $\mathcal{EL}^+$, that supports role inclusions. This approach can be implemented with low cost by leveraging any existing relational database management system to do the ABox data completion and query answering. We conducted a preliminary experiment to evaluate our approach using a large clinical data set and show our approach is practical.

The approach of using ontology reasoning to cleanse the output of information extraction tools was first articulated in SemantiClean. A limiting factor in applying this approach has been that ontology reasoning to find inconsistencies does not scale to the size of data produced by information extraction tools. In this paper, we describe techniques to scale inconsistency detection, and illustrate the use of our techniques to produce a consistent subset of a knowledge base with several thousand inconsistencies.

Reasoning on OWL ontologies is known to be intractable in the worst-case, which is a serious problem because in practice, most OWL ontologies have large Aboxes, i.e., numerous assertions about individuals and their relations. We propose a technique that uses a summary of the ontology (summary Abox) to reduce reasoning to a small subset of the original Abox, and prove that our techniques are sound and complete. We demonstrate the scalability of this technique for consistency detection in 4 ontologies, the largest of which has 6.5 million role assertions.

As XQuery is gathering momentum as the standard query language for XML, there is a growing interest in using it as an integral part of the XML application development infrastructure. In that context, one question which is often raised is how well XQuery interoperates with other XML languages, and notably with XSLT. XQuery 1.0 [16] and XSLT 2.0 [7] share a lot in common: they share XPath 2.0 as a common sub-language and have the same expressiveness. However, they are based on fairly different programming paradigms. While XSLT has adopted a highly declarative template based approach, XQuery relies on a simpler, and more operational, functional approach.In this paper, we present an approach to compile XSLT 2.0 into XQuery 1.0, and a working implementation of that approach. The compilation rules explain how XSLT's template-based approach can be implemented using the functional approach of XQuery and underpins the tight connection between the two languages. The resulting compiler can be used to migrate a XSLT code base to XQuery, or to enable the use of XQuery runtimes (e.g., as will soon be provided by most relational database management systems) for XSLT users. We also identify a number of areas where compatibility between the two languages could be improved. Finally, we show experiments on actual XSLT stylesheets, demonstrating the applicability of the approach in practice.

In order to make software components more flexible and reusable it is desirable to provide business users with facilities to assemble and control them, but without first being converted into programmers. We present our fully-functional prototype middleware system where variability is externalized so that core applications need not be altered for anticipated changes. Application behavior modification is fast and easy, suitable for a quickly changing e-commerce world.

Although the majority of evidence analysis in DeepQA is focused on unstructured information (e.g., natural-language documents), several components in the DeepQA system use structured data (e.g., databases, knowledge bases, and ontologies) to generate potential candidate answers or find additional evidence. Structured data analytics are a natural complement to unstructured methods in that they typically cover a narrower range of questions but are more precise within that range. Moreover, structured data that has formal semantics is amenable to logical reasoning techniques that can be used to provide implicit evidence. The DeepQA system does not contain a single monolithic structured data module; instead, it allows for different components to use and integrate structured and semistructured data, with varying degrees of expressivity and formal specificity. This paper is a survey of DeepQA components that use structured data. Areas in which evidence from structured sources has the most impact include typing of answers, application of geospatial and temporal constraints, and the use of formally encoded a priori knowledge of commonly appearing entity types such as countries and U.S. presidents. We present details of appropriate components and demonstrate their end-to-end impact on the IBM Watsoni system.

This paper presents the Mapping Manuscript Migrations (MMM) system in use for modeling, aggregating, publishing, and studying heterogeneous, distributed premodern manuscript databases on the Semantic Web. A general "Sampo model" is applied to publishing and using linked data in Digital Humanities (DH) research and to creating the MMM system that includes a semantic portal and a Linked Open Data (LOD) service. The idea is to provide the manuscript data publishers with a novel collaborative way to enrich their contents with related data of the other providers and by reasoning. For the end user, the MMM Portal facilitates semantic faceted search and exploration of the data, integrated seamlessly with data analytic tools for solving research problems in manuscript studies. In addition, the SPARQL endpoint of the LOD service can be used with external tools for customized use in DH research and applications. The MMM services are available online, based on metadata of over 220 000 manuscripts from the Schoenberg Database of Manuscripts of the Schoenberg Institute for Manuscript Studies (University of Pennsylvania), the Medieval Manuscripts in Oxford Libraries, and Bibale of Institut de recherche et d'histoire des textes in Paris. Evaluation of the MMM Portal suggests that the system is useful in manuscript studies and outperforms current systems online in searching, exploring, and analyzing data.