Anu Question Answering System

Abstract

AnuQA is a question answering system built on top of a search index and an enterprise knowledge graph. In this work, we describe ve semantic technologies that have helped us address real world challenges in deploying this system. These challenges include bias in knowledge base population, entity re-resolution on streaming data, ontology alignment across data sources, explaining relationships, and providing a single uni ed query interface for business analytics. [2] introduced the Anu Cognitive Compliance platform. It has enabled research in a number of elds including Search Index Optimization, Answer Sentence Selection, Document Similarity, Hypernym Discovery, Fine Grained Entity Classi cation, Ontology Creation and Link Prediction. We now present ve semantic technologies that we have implemented to enable real world deployments of AnuQA system.

-

Uni ed Hierarchical Label Set model for Ontology Alignment AnuQA requires fusing information from di erent data sources to enable natural language querying. This is typically handled by manual processes which become cumbersome as the number of sources increases. We use the Uni ed Hierarchical Label Set (UHLS) model [ 1 ], based on collective learning of entity types, to integrate labels from di erent data sources and standard ontologies.

Explainable Link Prediction

While a number of interpretability solutions have been proposed for link prediction by graph neural networks, human understandable explanations are desirable in real world applications. Based on [ 3 ], we extract supporting text from unstructured documents, logs, lineage data and relational tables. We also look at existing paths to explain new links predicted between nodes in our knowledge graph.

Reasoning for Natural Language Interpretation

Natural Language Query [ 4 ] interfaces allow end-users to ask questions without knowing any specialized query language or data storage and schema details. We use logical reasoning over domain semantics and knowledge to support a wide variety of domain-speci c queries in natural language. Domain reasoning helps us to make better interpretation of implicit intents in natural language queries, especially analytic queries typically posed to information access systems. Deployments Di erent parts of this question answering system have been deployed in various customer engagements and product o erings of IBM, especially in the nancial services domain. http://covid19-india-qa.mybluemix.net is a sample instance of the AnuQA system for answering questions on COVID19.

References

1. Abhishek , A. , Azad , A.P. , Ganesan , B. , Anand , A. , Awekar , A. : Collective learning from diverse datasets for entity typing in the wild . In: Proceedings of the 2nd International Workshop on EntitY REtrieval . pp. 16 { 23 . CEUR-WS ( 2019 )

2. Agarwal , A. , Ganesan , B. , Gupta , A. , Jain , N. , Karanam , H.P. , Kumar , A. , Madaan , N. , Munigala , V. , Tamilselvam , S.G. : Cognitive compliance for nancial regulations . IT Professional 19 ( 4 ), 28 { 35 ( 2017 )

3. Bhatia , S. , Dwivedi , P. , Kaur , A. : That's interesting, tell me more! nding descriptive support passages for knowledge graph relationships . In: International Semantic Web Conference . pp. 250 { 267 . Springer ( 2018 )

4. Sen , J. , Ozcan , F. , Quamar , A. , Stager , G. , Mittal , A. , Jammi , M. , Lei , C. , Saha , D. , Sankaranarayanan , K. : Natural language querying of complex business intelligence queries . In: Proceedings of the 2019 International Conference on Management of Data . pp. 1997 { 2000 ( 2019 )