Microsoft Technology Licensing, LLC

Gu Xu

Wei-Ying Ma

Hang Li

Zheng Chen

Jianfeng Gao

Benyu Zhang

Hua-Jun Zeng

Randy Hinrichs

Hongbin Gao

Hong-Jiang Zhang

Bin Shi

Qi Yao

Yu-Fei Ma

Bhuvan Middha

Jinxi Xu

Zhonghua Qu

Kyrylo Tropin

Jiafeng Guo

Ka Cheung Sia

Sandeep Dey

Jingfang Xu

Junbiao Tang

Systems, methods, and computer-readable storage media are provided for utilizing part-of-speech (POS) tagging of both the words included in a search query and the words included in potential search result documents to improve query alteration accuracy and search result ranking. Upon receiving a search query, POS tags are assigned to the words comprising the query to create query word-tag pairs. The query word-tag pairs are utilized to reformulate the query and are compared with document word-tag pairs included in a plurality of potential search result documents to determine a degree of similarity. The degree of similarity is utilized as an input in scoring and/or ranking the relevance of the potential search result documents with respect to one another.

Part-of-speech tagging for ranking search results

Candidate string transformations may be generated from applying applicable transformation rules from a rules index. Candidate string transformations may be eliminated based on a score associated with the candidate string transformation. The score may be based on probabilistic values associated with the applied applicable transformation rules.

Rule-based generation of candidate string transformations

Various technologies described herein pertain to use of path-constrained random walks for query expansion and/or query document matching. Clickthrough data from search logs is represented as a labeled and directed graph. Path-constrained random walks are executed over the graph based upon an input query. The graph includes a first set of nodes that represent queries included in the clickthrough data from search logs, a second set of nodes that represent documents included in the clickthrough data from the search logs, a third set of nodes that represent words from the queries and the documents, and edges between nodes that represent relationships between queries, documents, and words. The path-constrained random walks include traversals over edges of the graph between nodes. Further, a score for a relationship between a target node and a source node representative of the input query is computed based at least in part upon the path-constrained random walks.

Query expansion and query-document matching using path-constrained random walks

A method and system for identifying information about people is provided. The information system identifies groups of people that have relationships based on their relationships to documents or more generally to objects. The information system initially is provided with an indication of which people have which relationships to which documents. The information system then identifies clusters of people based on having a relationship to the same objects. The information system may also identify clusters of related objects associated with a cluster of people. When a user wants to identify information about a person, the user can provide the name of that person to the information system. The information system then can retrieve and display the names of the other people who are in the same cluster as the person.

Method and system for mining information based on relationships

A computer-implemented method and system for Web search ranking are provided herein. The method includes generating a number of training samples from clickthrough data, wherein the training samples include positive query-document pairs and negative query-document pairs. The method also includes discriminatively training a translation model based on the training samples and ranking a number of documents for a Web search based on the translation model.

Web search ranking

Named Entity Recognition in Query (NERQ) involves detection of a named entity in a given query and classification of the named entity into one or more predefined classes. The predefined classes may be based on a predefined taxonomy. A probabilistic approach may be taken to detecting and classifying named entities in queries, the approach using either query log data or click through data and Weakly Supervised Latent Dirichlet Allocation (WS-LDA) to construct and train a topic model.

Named entity recognition in query

Techniques are described for determining queries that are similar to rare queries. An n-gram space is defined to represent queries and a similarity function is defined to measure the similarities between queries. The similarity function is learned by leveraging training data derived from user behavior data and formalized as an optimization problem using a metric learning approach. Furthermore, the similarity function can be defined in the n-gram space, which is equivalent to a cosine similarity in a transformed n-gram space. Locality sensitive hashing can be exploited for efficient retrieval of similar queries from a large query repository. This technique can be used to enhance the accuracy of query similarity calculation for rare queries, facilitate the retrieval of similar queries and significantly improve search relevance.

Learning similarity function for rare queries

A crawling system crawls a web site initially in a pattern detection phase and subsequently in a pattern usage phase. The pattern detection phase attempts to identify patterns of references to pages that contain informational content of interest and patterns of references to pages that contain little informational content of interest. During the pattern usage phase, the crawling system crawls the web site. When the crawling system encounters a reference contained on an accessed page, the crawling system determines whether the reference matches a reference pattern. If the reference matches a reference pattern associated with pages that contain informational content of interest, the crawling system accesses the referenced page. If, however, the reference matches a reference pattern of pages with little informational content, then the crawling system discards that reference without accessing the referenced page.

Web forum crawler

Systems and methods for representing sequential motion patterns are described. In one aspect, video frames are converted into a sequence of energy redistribution measurements. One or more motion filters are then applied to the ER measurements to generate one or more temporal sequences of motion patterns, the number of temporal sequences being a function of the number of motion filters.

Growing community of inventors

Beijing, China

Gu Xu