Yahoo! Inc.

Jing Bai

Yi Chang

Zhaohui Zheng

Anlei Dong

Ruiqiang Zhang

Ciya Liao

Gilbert Leung

Fernando David Diaz

Karolina Buchner

Georges-Eric Albert Marie Robert Dupret

Gilad Avraham Mishne

Pranam Kolari

Shihao Ji

Ling Liu

In one embodiment, access a set of recency ranking data comprising one or more recency search queries and one or more recency search results, each of the recency search queries being recency-sensitive with respect to a particular time period and being associated with a query timestamp representing the time at which the recency search query is received at a search engine, each of the recency search results being generated by the search engine for one of the recency search queries and comprising one or more recency network resources. Construct a plurality of recency features from the set of recency ranking data. Train a first ranking model via machine learning using at least the recency features.

Incorporating recency in network search using machine learning

An information retrieval system is described herein that monitors a microblog data stream that includes microblog posts to discover and index fresh resources for searching by a search engine. The information retrieval system also uses data from the microblog data stream as well as data obtained from a microblog subscription system to compute novel and effective features for ranking fresh resources which would otherwise have impoverished representations. An embodiment of the present invention advantageously enables a search engine to produce a fresher set of resources and to rank such resources for both relevancy and freshness in a more accurate manner.

Ranking of search results based on microblog data

Embodiments are directed towards generating market-specific ranking models by leveraging target market specific pairwise preference data. The pairwise preference data includes market-specific training examples, while a ranking model from another market captures the common characteristics of the resulting ranking model. In one embodiment, the ranking model is trained by applying a Tree Based Ranking Function Adaptation (TRADA) algorithm to multi-grade labeled training data, such as editorially generated training data. Then, contradictions between the TRADA generated ranking model and target-market specific pairwise preference data are identified. For each identified contradiction, new training data is generated to correct the contradiction. Then, in one embodiment, an algorithm such as TRADA is applied to the existing ranking model and the new training data to generate a new ranking model.

Cross-market model adaptation with pairwise preference data

Growing community of inventors

San Jose, CA, United States of America

Jing Bai