Microsoft Technology Licensing, LLC

Shuang Peng

Jin Li

Li Zhou

Guang Yang

Jian H Li

Kun Jing

Xiaosong Yang

Min Zeng

Zhenyu Qian

Haoyong Zhang

Bowen Wu

Baoxun Wang

Non-limiting examples of the present disclosure describe a convolutional neural network (CNN) architecture configured to evaluate conversational relevance of query-response pairs. A CNN model is provided that can include a first branch, a second branch, and multilayer perceptron (MLP) layers. The first branch includes convolutional layers with dynamic pooling to process a query. The second branch includes convolutional layers with dynamic pooling to process candidate responses for the query. The query and the candidate responses are processed in parallel using the CNN model. The MLP layers are configured to rank query-response pairs based on conversational relevance.

Conversational relevance modeling using convolutional neural network

Entity pages are created that are optimized for search engines to return entity information from the entity pages in response to search queries. An entity page may be created for an entity by identifying electronic content sources that include data about the entity. Usage data indicative of how users have consumed data at the electronic content sources is also determined. The usage data is analyzed to identify topics for the entity and topic content is retrieved from the electronic content sources to create topic summaries. An entity page with the topics summaries is generated. When a search engine receives a search query related to the entity, the search engine may provide information from the entity page in response to the search query.

Entity page generation and entity related searching

An automated arrangement for parsing location descriptions is provided in which semantic verification is integrated into a parsing process to reduce the generation of false results. The semantic verification involves checking up to three semantic relationships between keywords (i.e., syntactical components) parsed from the location description in a tokenization process to determine if a tokenization result is valid. The semantic relationships include: a) a spatial 'part-of' relationship between location keywords; b) a spatial 'near-by' relationship; and, c) a spatial &#x201c;intersect&#x201d; relationship. The semantic relationships between particular locations may be pre-calculated and stored as extended vocabulary to enable the semantic verification to occur early in the parsing process to thus increase overall parsing efficiency. The results of the parsing are sorted based on a rank score that is derived using the semantic relationships between the locations.

Semantic relationship-based location description parsing

Growing community of inventors

Redmond, WA, United States of America

Shuang Peng