Кириллов Алексей Павлович
Automated Verification of Facts Using a Knowledge Base
Науки о данных
In our work we considered the problem of textual fact-checking using a knowledge base (KB). We analyzed state-of-the-art methods of checking triples (subject, predicate, object) using a KB and extracting such triples from a natural language text. Several algorithms for triples checking were implemented, such as the algorithm based on properties comparison, the algorithm which collects the nearest neighbors of a subject and an object using random walk on graph and uses collected data to measure the proximity between the subject and the object and the algorithm based on alternative paths detection – sequences of predicates which most frequently bind the same pair of entities as a target predicate. There was also developed the algorithm for preparation of data extracted from sentences for checking with KB. We collect several datasets of triples and a dataset of English sentences for implemented algorithms evaluation. The results of evaluation were analyzed with the weaknesses of algorithms being described. We have also come to a conclusion and described a plan for further work.
Текст работы (работа добавлена 22 мая 2019г.)