Intelligent SPARQL Query Generation for Natural Language Processing Systems

Abstract

Developing question answering (QA) systems that process natural language is a popular research topic. Conventionally, when QA systems receive a natural language question, they choose useful words or phrases based on their parts-of-speech (POS) tags. In general, words tagged as nouns are mapped to class entities, words tagged as verbs are mapped to property entities, and words tagged as proper nouns are mapped to named entities, although the accuracy of entity type identification remains low. Afterward, the relationship between entity types as RDF types determines the first element to be a pivot word to generate the SPARQL (acronym for SPARQL protocol and RDF query language) query on the basis of the sequences by a specific graph or tree structure, such as dependence tree or directed acyclic graph (DAG). However, the generated SPARQL query is difficult to adapt to the given query request in that the sequences are decided by a fixed structure. Unlike in previous research, SPARQL generation occurs automatically according to the entity type identification and RDF type identification results. This study attempts to design a method that leverages machine learning to learn human experiences in entity type identification as well as RDF-type identification. We approach the problem as a multiclass classification problem and propose a two-stage maximum-entropy Markov model (MEMM). The first stage identifies the entity type and the second identifies the RDF type for the purpose of generating appropriate SPARQL queries to meet the query request. Along with the templates designed for the two-stage MEMM model, we develop an automatic question answering prototype system called QAWizard. The experimental results show that QAWizard outperforms all other systems in question answering when evaluated on Linked Data version 8 (QALD-8) metrics.INDEX TERMS Question answering system (QA), Parts-of-speech (POS), SPARQL query, Maximumentropy Markov model.

SPARQL Generation with an NMT-based Approach

“…If the input sequence contains the word 'USA', the corresponding entity of this word in the target sequence should be 'dbr:United States'. Because 'USA' is now marked as 'NER', the post processing step can easily convert it to 'dbr:United States' based on the entity type tag [11,14]. However, it is very hard for any other translator to convert 'USA' to 'United States' and find the corresponding entity 'dbr:United States' due to the lexical gap and ambiguity problem.…”

“…The modification component is designed as a remedy for mistranslation problems. Since vocabulary mistranslation is profusely found in translation results, we employ entity type tagger from Chen et al Study [11] to help modify the result. Thus, the final SPARQL queries are decided jointly by both the result of Transformer model and the entity type tags.…”

“…In consideration of space, entity mapping will not be discussed in this paper. For readers who are interested in entity mapping, they can reference Chen et al Study [11].…”

“…The first one is to turn natural language questions into an intermediary format (ex. entity type tags), generally with some multiclass classifiers, followed by templates or dependency structure to generate the final SPARQL queries [10][11][12][13][14]. The second one is to rely on the structure of knowledge graph, applying algorithms for subgraph searching to find all possible RDF triples [15].…”

“…natural language questions), pre-trained word embedding is utilized to minimize the out-of-vocabulary problem. Another specialty of our presented approach is that we modify translation results by entity type tags [11,14] to help making up for the limited target vocabulary size.…”