Natural Language Conversion to SPARQL query

Semantic understanding of Chinese natural language questions in the Knowledge base

Objective: To convert the natural language of Chinese into SPARQL query

Basic methods:

Natural language preprocessing: Word segmentation (Ictclas), named entity recognition, syntactic analysis (Standford Parser) get a parse tree

1, using the syntactic analysis tree to construct the user's query semantic graph (Query semantic graph is a graph used to describe the entity relationship in the user's query, where vertices represent named entities or noun variables, and edges represent relationships between vertices)

2. Entity disambiguation (the vertices in the query semantics map are mapped to the entities in the Knowledge Base)

3. Predicate disambiguation (maps edges in query semantics to edges in the knowledge Base)

Map keywords in natural language to relationships in the knowledge base

3.1 Collection of keywords with high correlation (first, a relatively high correlation of verb phrases and noun phrases is collected for each relationship in the Knowledge Base)

3.1.1 For any relationship A, first extract the entity to R (a) <s,o> put the entity pair set P (a), collect all the sentences that appear in the entity pairs in P (a), put S (a), annotate the sentences in part-of-speech, and treat the set of nouns and verbs as a single document. The document computes the TF-IDF value of each word uniformly, and then sorts it in descending order. This gives a list of words that are related to each relationship and are sorted from high to low in terms of relevance.

3.2 Predicate mapping (defines an algorithm for calculating the relevance of words, using the collected words to map the predicates in the graph to the most semantically relevant relationships)

4 Convert a query semantic diagram to a SPARQL statement

Natural Language Conversion to SPARQL query