Microsoft Technology Licensing, LLC

Wen-tau Yih

Christopher A Meek

Joshua T Goodman

Jianfeng Gao

Christopher Brian Quirk

Kristina Nikolova Toutanova

Hoifung Poon

Li Deng

Geoffrey John Hulten

Xiaodong He

Aleksander R Kolcz

Lucretia Henrica Vanderwende

Hisami Suzuki

Paul Smolensky

Hai Wang

Vitor Rocha De Carvalho

Moontae Lee

Nanyun Peng

Hannaneh Hajishirzi

A technique is described herein for processing documents in a time-efficient and accurate manner. In a training phase, the technique generates a set of initial training examples by associating entity mentions in a text corpus with corresponding entity identifiers. Each entity identifier uniquely identifies an entity in a particular ontology. The technique then removes noisy training examples from the set of initial training examples, to provide a set of filtered training examples. The technique then applies a machine-learning process to generate a linking component based, in part, on the set of filtered training examples. In an application phase, the technique uses the linking component to link input entity mentions with corresponding entity identifiers. Various application systems can leverage the capabilities of the linking component, including a search system, a document-creation system, etc.

Distant supervision for entity linking with filtering of noise

A processing unit can determine multiple representations associated with a statement, e.g., subject or predicate representations. In some examples, the representations can lack representation of semantics of the statement. The computing device can determine a computational model of the statement based at least in part on the representations. The computing device can receive a query, e.g., via a communications interface. The computing device can determine at least one query representation, e.g., a subject, predicate, or entity representation. The computing device can then operate the model using the query representation to provide a model output. The model output can represent a relationship between the query representations and information in the model. The computing device can, e.g., transmit an indication of the model output via the communications interface. The computing device can determine mathematical relationships between subject representations and attribute representations for multiple statements, and determine the model using the relationships.

Computational-model operation using multiple subject representations

Long short term memory units that accept a non-predefined number of inputs are used to provide natural language relation extraction over a user-specified range on content. Content written for human consumption is parsed with distant supervision in segments (e.g., sentences, paragraphs, chapters) to determine relationships between various words within and between those segments.

Graph long short term memory for syntactic relationship discovery

Described is a technology for measuring the similarity between two objects (e.g., documents), via a framework that learns the term-weighting function from training data, e.g., labeled pairs of objects, to develop a learned model. A learning procedure tunes the model parameters by minimizing a defined loss function of the similarity score. Also described is using the learning procedure and learned model to detect near duplicate documents.

Learning element weighting for similarity measures

There is provided a computer-implemented method and system for ranking documents. The method includes identifying a number of query-document pairs based on clickthrough data for a number of documents. The method also includes building a latent semantic model based on the query-document pairs and ranking the documents for a search based on the latent semantic model.

Clickthrough-based latent semantic model

Two methods for measuring keyword-document relevance are described. The methods receive a keyword and a document as input and output a probability value for the keyword. The first method is a similarity-based approach which uses techniques for measuring similarity between two short-text segments to measure relevance between the keyword and the document. The second method is a regression-based approach based on an assumption that if an out-of-document phrase (the keyword) is semantically similar to an in-document phrase, then relevance scores of the in and out-of document phrases should be close to each other.

Consistent phrase relevance measures

Extraction analysis techniques biased, in part, by query frequency information from a query log file and/or search engine cache are employed along with machine learning processes to determine candidate keywords and/or phrases of web documents. Web oriented features associated with the candidate keywords and/or phrases are also utilized to analyze the web documents. A keyword and/or phrase extraction mechanism can be utilized to score keywords and/or phrases in a web document and estimate a likelihood that the keywords and/or phrases are relevant, for example, in an advertising system and the like.

Web document keyword and phrase extraction

The claimed subject matter provides systems and/or methods for normalizing document representations for use with Na&#xef;ve Bayes. The system can include devices and components that determine norms associated with documents by aggregating absolute term weight values associated with the documents, and further ascertain term weights for features associated with the documents, and thereafter divides the term weights for the features associated with the documents with the norms associated with the documents to produce a normalized document representation that can be utilized by arbitrary linear classifiers.

Raising the baseline for high-precision text classifiers

Document summarization is performed by scoring individual words in sentences in a document or document cluster. Sentences from the document or document cluster are selected to form a summary based on the scores of the words contained in those sentences.

Document summarization by maximizing informative content words

A system and method that facilitates and effectuates optimizing a classifier for greater performance in a specific region of classification that is of interest, such as a low false positive rate or a low false negative rate. A two-stage classification model can be trained and employed, where the first stage classification is optimized over the entire classification region and the second stage classifier is optimized for the specific region of interest. During training the entire set of training data is employed by a first stage classifier. Only data that is classified by the first stage classifier or by cross validation to fall within a region of interest is used to train the second stage classifier. During classification, data that is classified within the region of interest by the first classification is given the first stage classifier's classification value, otherwise the classification value for the instance of data from the second stage classifier is used.

Growing community of inventors

Redmond, WA, United States of America

Wen-tau Yih