Abstract: Search results are filtered using one or more rankers that evaluate a searcher intention to select items from a subset of search results. A subset of items may be provided that are responsive to a user search query. This subset of items may have one or more related properties and may be logically grouped together. A database of rankers may be evaluated and applied to the subset of items to determine items that are both relevant to the user search query and also correspond to a user intention for the search. As a result, the ranker may select and filter out certain items having one or more features that do not correspond to the user intention.

Abstract: System and methods are provided that can address cold-start problems in database keyword searches. The search system generates machine-learned values for new items based on historical signals for already existing items. These initial values are generated at the time of new item's inclusion in the search index. The values are used as input in a ranking model to rank search results for a user query. The initial values for the new items predict user engagement with the new items based on historical data for existing items and increase the visibility of new items to accumulate user interaction data for the new items.

Abstract: Embodiments of the present invention provide improved techniques for determining long term relevance and user behavior using query chains. The query chains may first be detected and then annotated into different types of chains based at least in part on various decision rules, machine-learned classifiers, and inter-query relationships. The query chains may then be subsequently used to train models for predicting user behavior and providing more relevant results to a user's queries. A content provider system according to various embodiments may aggregate historical data associated with previous search and/or transaction data, which may be analyzed to detect query chains, for example, whether queries are chained to each other. Determining whether queries are chained to each other may involve incorporating decision rules and reformulation models, analyzing temporal windows between queries, and/or analyzing inter-query relationships.

Abstract: Documents likely to be near-duplicates are clustered based on document vectors that represent word-occurrence patterns in a relatively low-dimensional space. Edit distance between documents is defined based on comparing their document vectors. In one process, initial clusters are formed by applying a first edit-distance constraint relative to a root document of each cluster. The initial clusters can be merged subject to a second edit-distance constraint that limits the maximum edit distance between any two documents in the cluster. The second edit-distance constraint can be defined such that whether it is satisfied can be determined by comparing cluster structures rather than individual documents.

Abstract: An automated parser for e-mail messages identifies component parts such as header, body, signature, and disclaimer. The parser uses a hidden Markov model (HMM) in which the lines making up an e mail are treated as a sequence of observations of a system that evolves according to a Markov chain having states corresponding to the component parts. The HMM is trained using a manually-annotated set of e-mail messages, then applied to parse other e-mail messages. HMM-based parsing can be further refined or expanded using heuristic post-processing techniques that exploit redundancy of some component parts (e.g., signatures, disclaimers) across a corpus of e-mail messages.

Abstract: An automated parser for e-mail messages identifies component parts such as header, body, signature, and disclaimer. The parser uses a hidden Markov model (HMM) in which the lines making up an e mail are treated as a sequence of observations of a system that evolves according to a Markov chain having states corresponding to the component parts. The HMM is trained using a manually-annotated set of e-mail messages, then applied to parse other e-mail messages. HMM-based parsing can be further refined or expanded using heuristic post-processing techniques that exploit redundancy of some component parts (e.g., signatures, disclaimers) across a corpus of e-mail messages.

Abstract: Documents likely to be near-duplicates are clustered based on document vectors that represent word-occurrence patterns in a relatively low-dimensional space. Edit distance between documents is defined based on comparing their document vectors. In one process, initial clusters are formed by applying a first edit-distance constraint relative to a root document of each cluster. The initial clusters can be merged subject to a second edit-distance constraint that limits the maximum edit distance between any two documents in the cluster. The second edit-distance constraint can be defined such that whether it is satisfied can be determined by comparing cluster structures rather than individual documents.