Computer-generated sentiment-based knowledge base

摘要

A computer-automated method and system of providing a searchable knowledge base with decision-relevant attributes (including some subjective or sentiment-based attributes) for a plurality of individual items within a choice set are described. First, information (including texts) relevant to the plurality of individual items in the choice set is harvested from Internet sources. Next, normalized representations of statements are extracted from excerpts of the harvested texts that pertain to attributes of interest for the choice set, and corresponding scores for the attributes are derived from each of the normalized representations. The scores derived from the various harvested sources are aggregated for each attribute of each item. Finally, the knowledge base of the plurality of individual topics is generated.

主权项

1. A computer-automated method of generating a searchable knowledge base for a plurality of items in a choice set, each of said items having an associated set of attributes including at least one subjective attribute, comprising the steps of:
harvesting information relevant to each of said items as well as a source and an age of said harvested information from the Internet; extracting a set of normalized representations from a corresponding set of statements in one or more texts of said harvested information; analyzing the normalized representations to derive a set of scores for each attribute in said associated set of attributes for each of said items; aggregating said derived scores, for said each attribute in said associated set of attributes, by assigning a weight to each of the scores based on at least the source and the age of said harvested information; and incorporating attributes in said associated set of attributes and their corresponding aggregated scores, for each of said items, into said searchable knowledge base for said choice set, wherein a computer processor is utilized to automatically perform each of steps, the incorporating including:
annotating a first statement harvested for a first item of the plurality of items with a first set of attributes and a corresponding first set of scores to generate a first annotation, the first set of attributes associated with the first item and determined based on an analysis of text in the first statement, the first statement being one of multiple statements harvested for the first item, the first annotation being one of multiple annotations in the searchable knowledge base, the first annotation including a link to the first statement, andindexing the annotations in the searchable knowledge base to facilitate retrieval of a statement of the statements that corresponds to an input attribute.