Oracle International Corporation

Walmart Apollo, LLC

Yahoo! Inc.

Microsoft Technology Licensing, LLC

Ciya Liao

Thomas Chang

Meeten Bhavsar

Muralidhar Krishnaprasad

Yi Chang

Zhaohui Zheng

Joaquin Delgado

Anlei Dong

Georges-Eric Albert Marie Robert Dupret

Leonardo Lezcano

Omar Rogelio Alonso

Cindy Hsin

Hiroshi Koide

Narayanan Sadagopan

Shamim A Alpha

Pankaj Appasaheb Adsul

Jorge Gomez Robles

Jesus Perez Martin

Krishna Sravanthi Rajanala Sai

Yoshiyuki Inagaki

Jie Cai

Thomas Baby

Sachin Bhatkar

Visar Nimani

Mark Ture

Hui Ouyang

Gilbert Leung

Mark Davis

Karolina Buchner

Gilad Avraham Mishne

Jing Bai

Rajyashree Mukherjee

Ruiqiang Zhang

Vachik Shailesh

Krishna Sravanthi

Steve Chi-Ming Yang

Kaushal Yagnik

Tony Eilo Lee

Satya Chembolu

Ling Liu

Shihao Ji

A method can include determining a translatability class label and a class-associated search query for a cross-lingual search query. Determining the translatability class label and the class-associated search query can include determining whether a first class-translation pair for the cross-lingual search query exists in a memory module. If the first class-translation pair exists in the memory module, the method further can include retrieving the first class-translation pair for the cross-lingual search query from the memory module. When no class-translation pair for the cross-lingual search query exists in the memory module, the method additionally can include: determining the translatability class label, and determining the class-associated search query based on the translatability class label. The method further can include determining search results for the cross-lingual search query based on the translatability class label and the class-associated search query, and transmitting the search results to a user based on the translatability class label. Other embodiments are disclosed.

System and method for performing cross-lingual product searches

A method can include determining, via a translatability classifier module, a translatability class label for a cross-lingual search query received, via a computer network, from a user device for a user. The translatability classifier module can be trained to determine the translatability class label among multiple translatability class labels associated with user intentions for the cross-lingual search query based on a respective probability of the cross-lingual search query being associated with each of the multiple translatability class labels. The method further can include determining, via a language translator module, a class-associated search query for the cross-lingual search query based on the translatability class label, as determined. The method additionally can include transmitting, via the computer network, the class-associated search query to a monolingual search engine. Other embodiments are disclosed.

System and method for processing cross-lingual search queries

Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and execute functions comprising: storing historical session data pertaining to user sessions and archived search queries submitted by users to a search engine; analyzing the historical session data to identify ambiguous queries, including semantically ambiguous queries and multi-language queries; monitoring search queries submitted to the search engine to detect the ambiguous queries; and in response to detecting an ambiguous query, generating a query resolution interface that displays categorical groupings, each of which corresponds to a possible intention of the ambiguous query. Other embodiments are disclosed herein.

Systems and methods for detecting and resolving ambiguous search queries

Systems and methods including one or more processors and one or more non-transitory storage devices storing computing instructions configured to run on the one or more processors and execute functions comprising: receiving, at a search engine, a search query comprising tokens; determining, using a weight determination model, token weights for the tokens included in the search query; comparing the token weights associated with the tokens to a probability threshold; identifying the tokens that satisfy the probability threshold; and utilizing the tokens that satisfy the probability threshold to identify the search results. Other embodiments are disclosed herein.

Systems and methods for determining and utilizing search token importance using machine learning architectures

A fast browsing architecture for exploring hierarchical lists of entities through a search user interface. A graphical UI operates to handle the hierarchical lists and sub-lists in different ways for different scenarios such as a hierarchical level is zero (only one list of entities associated with a query and the list cannot be further drilled down), a second scenario where the hierarchical level is one (a list of entities associated with the query and these entities can be further drilled down to a number of sub-lists) and the sub-lists cannot be further drilled down, and a third scenario where the hierarchical level is more than one (a list of entities associated with the query and these entities can be further drilled down to a number of sub-lists), sub-lists can be further drilled down to a number of lists, until there is no more drill down lists to be found.

Hierarchical entity information for search

A fast browsing architecture for exploring hierarchical lists of entities through a search user interface. A graphical UI operates to handle the hierarchical lists and sub-lists in different ways for different scenarios such as a hierarchical level is zero (only one list of entities associated with a query and the list cannot be further drilled down), a second scenario where the hierarchical level is one (a list of entities associated with the query and these entities can be further drilled down to a number of sub-lists) and the sub-lists cannot be further drilled down, and a third scenario where the hierarchical level is more than one (a list of entities associated with the query and these entities can be further drilled down to a number of sub-lists), sub-lists can be further drilled down to a number of lists, until there are no more drill down lists to be found.

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

A web crawler indexes documents including information about document contents and metadata including information such as a URL. However, some applications rely on URL's that change frequently or are constructed to include user information so that the contents retrieved is customized to the user. An approach is provided for storing generic URL's in an index at crawl time, which are customized for the user at search time. A callback mechanism may be used to dynamically transform the generic URL into a URL that is specific to the user issuing the query and/or includes current information that may change frequently. In this way, when the query or search results are returned to the user, the user receives links that are active and valid for that particular user, directing the user to the appropriate site, application, etc. without requiring continuous updating of a very large index.

Indexing secure enterprise documents using generic references

It is desirable to provide a secure search mechanism to provide for searching over any and all content, such as across an enterprise. A secure search, however, requires access to the secure content repositories holding the data to be searched. In some cases the credentials required to crawl a repository may be extremely sensitive, or the user may be reluctant or unwilling to store user identification information in memory or on disk for any longer than is absolutely necessary. An approach is provided that allows a user or an administrator to provide security credentials to be stored and used only during a crawl, and to erase the credentials from the system when the crawl is complete.

Crawling secure data sources

An enterprise-wide query relaxative support vector machine ranking algorithm approach provides enhanced functionality for query execution in a heterogeneous enterprise environment. Improved query results are obtained by adjusting ranking functions using machine learning methods to automatically train ranking functions. The improved query results are obtained using a list of document-query pairs that are modeled as a binary classification training problem, combination function which requires ranking and learning functions to be implemented representing document attributes and metadata utilizing query relaxation techniques and adjusted ranking functions. Machine learning methods implement user feedback to automatically train ranking functions.

System and method for applying ranking SVM in query relaxation

A flexible and extensible architecture allows for secure searching across an enterprise. Such an architecture can provide a simple Internet-like search experience to users searching secure content inside (and outside) the enterprise. The architecture allows for the crawling and searching of a variety or sources across an enterprise, regardless of whether any of these sources conform to a conventional user role model. The architecture further allows for security attributes to be submitted at query time, for example, in order to provide real-time secure access to enterprise resources. The user query also can be transformed to provide for dynamic querying that provides for a more current result list than can be obtained for static queries.

Link analysis for enterprise environment

Search term ranking algorithms can be generated and updated based on customer settings, such as where a ranking algorithm is modeled as a combination function of different ranking factors. An end user of a search system provides personalized preferences for weighted attributes, generally or for a single instance of the query. The user also can indicate the relative importance of one or more ranking factors by specifying different weights to the factors. Ranking factors can specify document attributes, such as document title, document body, document page rank, etc. Based on the attribute weights and the received user query, a ranking algorithm function will produce the relevant value for each document corresponding to the user preferences and personalization configurations.

Changing ranking algorithms based on customer settings

A flexible and extensible architecture allows for secure searching across an enterprise. Such an architecture can provide a simple Internet-like search experience to users searching secure content inside (and outside) the enterprise. The architecture allows for the crawling and searching of a variety of sources across an enterprise, regardless of whether any of these sources conform to a conventional user role model. The architecture further allows for security attributes to be submitted at query time, for example, in order to provide real-time secure access to enterprise resources. The user query also can be transformed to provide for dynamic querying that provides for a more current result list than can be obtained for static queries.

Secure search performance improvement

In one embodiment, access one or more query chains, wherein each one of the query chains comprises two or more search queries, {q, . . . , q}, which are recency-sensitive, are related to the same subject matter, and are issued to a search engine sequentially, and actual click-through information associated with each one of the query chains; and smooth each one of the query chains using the actual click-through information associated with the query chain. To smooth one of the query chains comprises, for each one of search queries, q, in the query chain, where 2&#x2266;j&#x2266;n, if one of the network resources identified for qhas actually been clicked in connection with qby the corresponding one network user, then presume that the one network resource has been clicked in connection with one or more search queries, q, in the query chain, where 1&#x2266;k&#x3c;j.

Session based click features for recency ranking

Acronym and synonym pairs can be identified and retrieved automatically in a corpus and/or across an enterprise based on customer settings globally or for a single instance. Possible acronym and synonym term pairs can be identified using a rule such as a heuristic, user-defined rule. Rules selected by the user can be used to rank acronym and synonym pairs using factors such as occurrence frequency and maximum term length. A rule interpreter engine executes the user defined rule set to properly identify and retrieve the user selected acronym and synonym pairs through the utilization of a shallow pause read step. Finally, the user selected acronym and synonym pairs are ranked according to the user preferences, and can be displayed or held for subsequent use in searching.

Growing community of inventors

Fremont, CA, United States of America

Ciya Liao