U S West, Inc.

Mediaone Group, Inc.

University Technology Corporation

U S West Technologies, Inc.

Thomas K Landauer

Douglas A Corey

Peter William Foltz

Walter Kintsch

Robert Ernest Rehder

Bud C Wonsiewicz

Robert Darrell Laham, Ii

Karen E Lochbaum

The present invention is a methodology for analyzing and evaluating a sample text, such as essay(s), or document(s). This methodology compares sample text to a reference essay(s), document(s), or text segment(s) within a reference essay or document. The methodology analyzes the amount of subject-matter information in the sample text, analyzes the relevance of subject matter information in the sample and evaluates the semantic coherence of the sample. This methodology presumes there is an underlying, latent semantic structure in the usage of words. The method parses and stores text objects and text segments from the sample text and reference text into a two-dimensional data matrix. A weight is computed for each text object and applied to each data matrix cell value. The method performs a singular value decomposition on the data matrix, which produces three trained matrices. The method computes a vector representation of the sample text and reference text using the three trained matrices. The methodology compares the sample text to the reference text by computing the cosine between the vector representation of the sample text and the vector representation of the standard reference text. Alternatively, the dot product is used to compare the sample text to the standard reference text. A grade is assigned to the sample text based on the degree of similarity between the sample text and the standard reference text.

Methods for analysis and evaluation of the semantic content of a writing based on vector length

An information retrieval system is disclosed, wherein the system includes a plurality of text search engines based on substantially different computational searching techniques. By activating each search engine with input from a user information request, the output from each of the search engines is combined into a single list of information items. A ranking process ranks the information items in the combined list by utilizing information item ordering data also received from each of the search engines as to the relevance of the information items output by the search engine to the user's request. Thus, by providing higher rankings to those information items determined to be most relevant to the user's request by each of (or a majority of) the search engines, these information items have been found to be highly consistent in satisfying the user's request for information.

Searching large collections of text using multiple search engines

The present invention is a system and method for retrieving segments of stored video programs using closed caption text data. The closed caption text data is extracted from video programming signals received by the invention. Text records based on the extracted closed caption data are generated. Each text record is derived from the closed caption data for a single continuous video segment to which the text record serves as an index or key in retrieving this video segment. Preferably, each text record (a) has sufficient content to adequately describe the content of the video segment to which it serves as an index; and (b) corresponds to a video segment focused on a small number of topics. To accomplish (a) and (b) the present invention generates each text record so that it has a predetermined maximum length and so that it is derived from the closed caption data for a single uninterrupted speaker. During video data retrieval, video requests or queries input by users are evaluated either by comparing query terms with terms in the text records, or by comparing an interpretation of query terms with an interpretation of terms in the text records. Subsequently, video segment location information associated with each text record satisfying a query is used to retrieve the video segment to which the text record is an index.

Video programming retrieval using extracted closed caption data which

Growing community of inventors

Boulder, CO, United States of America

Thomas K Landauer