Salesforce Com, Inc.

Emc Corporation

Xerox Corporation

Bruno Roustant

Pierre-Yves Chevalier

Ilan Ginzburg

Pierre Salagnac

Juan Camilo Rodriguez Duran

Satyapal P Reddy

Marc Brette

Christian Posse

Francois Lopitaux

Son Thanh Chang

Ghislain Brun

Jean-Baptiste Charlety

Xavier Pavard

Yves Mahe

A minimal distinguishing prefix (MDP) is computed for terms included in a list of terms. The MDP for a term is the minimal prefix that differentiates the term from the immediately preceding term. Lengths of the computed MDPs are compared, and a set of MDPs is selected for inclusion in a trie. A trie generator generates a trie including a plurality of leaf nodes based on the selected MDPs, where each leaf node in the trie corresponds to a respective one of the selected MDPs, such that the terms accessible through a particular leaf node include the term corresponding to the respective one of the selected MDPs and any other terms positioned in the list between that term and a term corresponding to a next one of the selected MDPs in the trie.

Indexing hems in a database environment

An entries array having a pre-selected number of initially empty buckets and structured as a circular array is maintained. Each of the pre-selected number of entries in the entries array has a corresponding index value. A chaining array having the pre-selected number of buckets and structured as a circular array is also maintained. Each of the pre-selected number of entries in the chaining array has the same corresponding index value as the corresponding bucket in the entries array. A hash operation is performed on data to be stored to determine a hash value corresponding to the data. The data is stored in a bucket in the entries array corresponding to the hash value as an unmovable head-of-chain entry, and moving previous data, if any, stored in the entries array corresponding to the hash value if the previous data has a different hash value. The data is stored in another bucket in the entries array if the bucket in the entries array corresponding to the hash value is occupied by a head-of-chain entry. The bucket storing the data is linked to the head-of-chain entry with the chaining array.

Worm hashing

A minimal distinguishing prefix (MDP) is computed for terms included in a list of terms. The MDP for a term is the minimal prefix that differentiates the term from the immediately preceding term. Lengths of the computed MPDs are compared, and a set of MDPs is selected for inclusion in a trie. A trie generator generates a trie including a plurality of leaf nodes based on the selected MDPs, where each leaf node in the trie corresponds to a respective one of the selected MDPs, such that the terms accessible through a particular leaf node include the term corresponding to the respective one of the selected MDPs and any other terms positioned in the list between that term and a term corresponding to a next one of the selected MDPs in the trie.

Indexing items in a database environment

A method for presenting search results is provided. The method receives a user input search query; obtains, from a search engine, a first set of search results responsive to the user input search query; efficiently identifies a promoted set of search results for promotion using a set of search promotion rules, each of the set of search promotion rules including a set of terms and one or more document identifiers; revises the first set of search results, based on the identified promoted set, to create a second set of search results; and presents the second set of search results, wherein the second set of search results includes the first set of search results and the promoted set in a promoted position.

Search promotion systems and method

A system associates sequentially descending weights with items in a reference list. The system identifies overlapping items from the first position to the second position in both the reference list and a test list. The system determines a weighted overlap for the overlapping items by combining the weights for each overlapping item to create a reference item weight, combining the weights for each item from the first position to the second position in the reference list to create a reference position weight, and determining the ratio between the reference item weight and the reference position weight. The system identifies overlapping items from the first position to the third position in both the reference list and the test list, and then determines a weighted overlap for these overlapping items. The system outputs the average weighted overlap as an efficiency evaluation of the information retrieval system that created the test list.

Asymmetric rank-biased overlap

A database system may generate a new case associated with a customer support query. The database system may initiate a search for other cases related to the new case and identify articles linked to the related cases. The database system may identify more relevant articles by finding the related cases that use a similar vocabulary for describing similar customer problems. The database system then identifies the articles used for resolving the prior problems. Thus, the database system may bridge the gap between vocabularies used by customers to describe problems and vocabularies used in articles to describe solutions to those problems.

Identifying relevant material for cases

Semantic-based processing techniques are disclosed. Semantic processing of content data comprising a content item is performed. A semantic processing-triggered event is triggered based at least in part on the semantic processing. A responsive action is performed in response to the event. The responsive action may include launching a business process, providing a result of semantic processing as an input, or augmenting a content item metadata set with a result of semantic processing.

Semantic-based business events

Phrase matching processes for matching phrases comprising a plurality of keywords in document text construct hit lists of the keywords in a document text, and operate on the keywords in either phrase order or without regard to the order of occurrence of the keywords in the phrase. The processes form sorted sets of all keywords, and compare occurrences of the keywords in the sorted sets to a predefined proximity constraint. For unordered phrases, the proximity constraint defines a maximum span between keywords in the highest and lowest positions in the sorted set as MaxSpan=p(k&#x2212;1), where p is a proximity and k is the number of keywords in the phrase. For ordered phrases, the distances between successive phrase keywords in phrase order must be less than or equal to the proximity p.

Phrase matching for document classification

The invention is a system and method for teaching and using a computer to apply a genetic algorithm in order to generate rules for classifying data objects, such as documents. By permitting periodic human intervention during and/or after the initial training stage, the invention improves upon existing methods that merely automate data classification. Using an initial 'training set' of data objects, the system develops a set of rules to apply on future data objects. To improve efficacy over time, rules may be periodically amended by a user during the initial training stage or after rules have been generated. In this fashion, classification rules may evolve, thereby improving the accuracy and speed of the classification process. Classification rules are also presented in a human-readable format, thereby improving the transparency of the classification process and facilitating the interaction between the system and the administrator.

Interactive semi-supervised machine learning for classification

Determining a relevancy ranking score is disclosed. A query is received that includes one or more constraints. One of the one or more constraints includes an operator. A search result based on the query is received. The relevancy ranking score for the received search result is determined based at least in part on the operator associated with the one or more constraints of the query.

Relevancy scoring using query structure and data structure for federated search

Determining a relevancy ranking score is disclosed. An indication is received that a relevancy ranking score algorithm is to be tuned to a selected preference. The relevancy ranking score algorithm is updated based at least in part on the selected preference, wherein the relevancy ranking score of a search result resulting from a search query is based at least in part on one or more constraints of the search query.

Growing community of inventors

Froges, France

Bruno Roustant