Microsoft Technology Licensing, LLC

Cheng Niu

Ming Zhou

Lei Shi

Ying Li

Li Erran Li

Hsiao-Wuen Hon

Zhaohui Tang

Brian Burdick

Chin-Yew Lin

Systems, methods, and computer-readable media for dynamically generating text associated with an advertisement are provided. Core text associated with an advertisement is received from an advertiser, as is at least one attribute relevant to the advertiser and/or a user. Based upon the received attribute(s), it is determined whether customization of the core text is desired. If customization is desired, the core text is modified and presented in association with the advertisement. If customization is not desired, the core text is presented in association with the advertisement. In one embodiment, target advertisement placement information may also be utilized to determine whether customization of the core text is desired.

Dynamic generation of advertisement text

A classifier is built to rank documents of different languages found in a query based at least in part on similarity to other documents and the relevance of those other documents to the query. A joint ranking model, e.g., based upon a Boltzmann machine, is used to represent the content similarity among documents, and to help determine joint relevance probability for a set of documents. The relevant documents of one language are thus leveraged to improve the relevance estimation for documents of different languages. In one aspect, a hidden layer of units (neurons) represents clusters (corresponding to relevant topics) among the retrieved documents, with an output layer representing the relevant documents and their features, and edges representing a relationship between clusters and documents.

Joint ranking model for multilingual web search

A set of candidate documents, each of which may be part of a bilingual, parallel set of documents, are identified. The set of documents illustratively includes textual material in a source language. It is then determined whether parallel text can be identified. For each document in the set of documents, it is first determined whether the parallel text resides within the document itself. If not, the document is examined for links to other documents, and those linked documents are examined for bilingual parallelism with the selected documents. If not, named entities are extracted from the document and translated into the target language. The translations are used to query search engines to retrieve the parallel correspondent for the selected documents.

Identifying parallel bilingual data over a network

A set of candidate parallel pages is identified based on trigger words in one or more pages downloaded from a given network location (such as a website). A set of document trees representing each of the candidate pages are aligned to identify translationally parallel content and hyperlinks. The parallel content is further fed into conventional sentence aligner for parallel sentences. And the parallel hyperlinks usually refer to other parallel documents, and lead to a recursive mining of parallel documents.

Aligning hierarchial and sequential document trees to identify parallel data

Cross-lingual query suggestion (CLQS) aims to suggest relevant queries in a target language for a given query in a source language. The cross-lingual query suggestion is improved by exploiting the query logs in the target language. CLQS provides a method for learning and determining a similarity measure between two queries in different languages. The similarity measure is based on both translation information and monolingual similarity information, and in one embodiment uses both the query log itself and click-through information associated therewith. Monolingual and cross-lingual information such as word translation relations and word co-occurrence statistics may be used to estimate the cross-lingual query similarity with a discriminative model.

Cross-lingual query suggestion

Cross-lingual search re-ranking is performed during a cross-lingual search in which a search query of a first language is used to retrieve two sets of documents, a first set in the first language, and a second set in a second language. The two sets of documents are each first ranked by the search engine separately. Cross-lingual search re-ranking then aims to provide a uniform re-ranking of both sets of documents combined. Cross-lingual search re-ranking uses a unified ranking function to compute the ranking order of each document of the first set and the second set of documents. The unified ranking function is constructed using generative probabilities based on multiple features, and can be learned by optimizing weight parameters using a training corpus. Ranking SVM algorithms may be used for the optimization.

Growing community of inventors

Beijing, China

Cheng Niu