Building a knowledge base system from unstructured text is one of challenges in natural language processing (NLP). A knowledge base can be applied to various artificial intelligence areas (e.g. expert system, big data processing). Slot filling is one way to build a knowledge base from unstructured text. We are given one entity, query (was called slot in slot-filling) and text can include information about answer of query.
For example, slot filling problem give one entity:Hank Williams, and query (slot):cause of death, and text:(The coroner’s report said Williams died of a “heart condition and hemorrhaging.”) Then we need to find that answer, cause of death of Hank Williams, is “heart condition and hemorrhaging.”
Slot-Filling is hard problem because we will need to find which sentence justify answer and solve anaphora problem. At Text Analysis Conference (TAC) 2014 best precision was 63.45, it is around half of human performance.(Mihai Surdeanu 2013) Some research team obtain excellent result using distant supervision with rules (Angeli et al. 2013). There are other method, building system based on OpenIE (Soderland et al. 2013) and bootstrapping dependency-based patterns (Li et al. 2013).
In this paper, our system was divided two part, find slot filler and check whether this filler is correct or not. Firstly using other knowledge base (e.g. freebase) extract feature of sentence from unstructured text. Before extraction, unstructured text is resolved anaphora problem. Then modify extracted feature. Lastly, we make some sentences using query and answer (The cause of death of Hank Williams is heart condition and hemorrhaging.). Then use Recognizing Textual Entailment (RTE) for checking whether given text can entail this sentence or not.