Knowledge Base Question Answering Clause Samples

Knowledge Base Question Answering. Despite the abundance of knowledge available in textual resources, it is often challenging for a computer system to extract and understand this informa- tion. Knowledge bases, on the other hand, encode precise factual information, which can be effectively queried and reasoned with, which is quite natural in computer science. In the early days of QA research, knowledge bases were relatively small and contained information specific to a particular domain. Many approaches have been developed to answer detailed questions about these domains, e.g., baseball [81], lunar geology [223], or geography [243]. Recent development of large scale knowledge bases (e.g., DBpedia [15], Freebase [34], YAGO [184], WikiData5) shifted attention towards open domain question answering. One of the problems of techniques developed in the earlier days is domain adaptation, as it is quite challenging to map from natural language phrases to database concepts in open domain when the search space is quite large. KBQA approaches can be evaluated on an annual Question Answering over 5▇▇▇▇://▇▇▇.▇▇▇▇▇▇▇▇.▇▇▇ Linked Data (QALD6) shared task, and some popular benchmark dataset, such as Free917 [46], WebQuestions [22] and WebQuestionsSP [240]. A series of QALD evaluation campaigns has started in 2011, and since then a number of different subtasks have been offered, i.e., since 2013 QALD includes a multilingual task, and QALD-4 formulated a problem of hybrid question answering. These tasks usually use DBpedia knowledge base and provide a training set of questions, annotated with the ground truth SPARQL queries. The hybrid track is of particular interest to the topic of this dissertation, as the main goal in this task is to use both structured RDF triples and free form text available in DBpedia abstracts to answer user questions. A survey of some of the proposed approaches can be found in [197].