Extraction and indexing of triplet-based knowledge using natural language processing
Hooge, David Carl
MetadataShow full item record
A proper understanding of any document relies heavily upon two things: an understanding of the relationships between terms and a grasp of the manner in which language relates one term to another. For example a full comprehension of the sentence “Jane plays basketball” requires the reader to first understand that Jane is related to basketball by her taking part in this activity; second, the reader must have an understanding that of how basketball relates to other terms. Thus, for a full grasp of the sentence the reader must be aware that basketball is a sport among other things. These two understandings are missing from current search and storage methodologies and are instead largely replaced with word distance measures. As such the only relation stored by most modern methods is that the word “Jane” appears near the word “basketball.” Our system remedies these two problems through both relationship recognition as well as a grasp of how concepts relate to one another as in the linking of “sports” to “basketball.” This allows for automated semantic information storage and beyond this enables storage of information in a manner that resembles the structure of language.