| CPC G06F 16/90332 (2019.01) [G06F 16/9024 (2019.01); G06F 16/93 (2019.01)] | 17 Claims |
|
1. A computer-implemented method for maximizing answer coherence of a multi-part query using extracted relationships between chunks of at least one support document, the computer-implemented method comprising:
receiving, by a processor, the multi-part query from a user;
identifying, by the processor, a number of parts comprising the multi-part query using natural language processing (NLP) based on identifying a tone or sentiment of the user associated with the multi-part query, wherein each of the number of parts corresponds to a list of potential candidate answers extracted as the chunks from the at least one support document;
creating, by the processor, a query graph comprising a plurality of answer nodes, wherein each of the answer nodes correspond to each of the potential candidate answers corresponding to each of the number of parts comprising the multi-part query, and edges connect a root node to answer nodes corresponding to a first part of the multi-part query, and the answer nodes corresponding to the first part of the multi-part query are connected by the edges to answer nodes corresponding to a subsequent part of the multi-part query;
forming, by the processor, a plurality of answer node pathways within the query graph, extending from the root node through each of answer nodes, wherein each answer node pathway follows a different set of answer nodes and edges connected along the query graph;
computing, by the processor, an answer coherence score for each answer node pathway, wherein computing the answer coherence score for each answer node pathway includes traversing, by the processor, each answer node pathway starting from the root node and for each edge between the answer nodes labelled as having a relationship between the answer nodes penalizing the edge, while rewarding edges between unrelated answer nodes along the answer node pathway, and wherein the answer coherence score for each answer node pathway penalizes edges along an answer node pathway that represent relationships between answer nodes and rewards answer node pathways that include edges between answer nodes that are unlabeled and unrelated to one another, and wherein the relationships correspond to nodes that are complementary to, contained within, or an instance of one another; and
selecting, by the processor, an answer to the multi-part query using the answer nodes along the answer node pathway having a maximum answer coherence score.
|