Abstract: Systems and methods for predictive document coding using continuous active machine learning are described herein. A method uses both a primary queue and a plurality of secondary queues, where each secondary queue is associated with a model for category of documents. The method also repeatedly classifies new batches selected from a large set of documents that have not been reviewed. The classification uses the plurality of models and updates the secondary queues from the best documents in the most recently classified batch. While the method transparently cycles through batches, the most relevant documents are provided to one or more human reviewers from secondary queues via a primary queue. The reviewer confirms relevance or non-relevance in each of the documents for each of the categories. Periodically all the models are retrained using the set of reviewed documents after a selectable number of documents have been reviewed since the most recent retraining.

Abstract: Systems and methods for improving search results from proprietary search engine technologies and proprietary machine classifiers, without ingesting, copying, or storing, the data to be searched, are described herein. A user sends a query to a proprietary search engines and gets a result document set back. The user may apply a user model for classifying a result document set to generate a result document for review of a user. The reviewed document may be added to a user training corpus, which is then used to retrain the user model. The retrained user model may be applied by the user to generate the next result document for user review and so on until the user model converges to generate relevant documents reliably. Once the user model converges, the user may apply the now reliable user model to generate multiple relevant documents for the user.

Abstract: Systems and methods for improving search results from proprietary search engine technologies and proprietary machine classifiers, without ingesting, copying, or storing, the data to be searched, are described herein. A user sends a query to a proprietary search engines and gets a result document set back. The user may apply a user model for classifying a result document set to generate a result document for review of a user. The reviewed document may be added to a user training corpus, which is then used to retrain the user model. The retrained user model may be applied by the user to generate the next result document for user review and so on until the user model converges to generate relevant documents reliably. Once the user model converges, the user may apply the now reliable user model to generate multiple relevant documents for the user.

Abstract: Systems and methods for predictive document coding using continuous active machine learning are described herein. A method uses both a primary queue and a plurality of secondary queues, where each secondary queue is associated with a model for category of documents. The method also repeatedly classifies new batches selected from a large set of documents that have not been reviewed. The classification uses the plurality of models and updates the secondary queues from the best documents in the most recently classified batch. While the method transparently cycles through batches, the most relevant documents are provided to one or more human reviewers from secondary queues via a primary queue. The reviewer confirms relevance or non-relevance in each of the documents for each of the categories. Periodically all the models are retrained using the set of reviewed documents after a selectable number of documents have been reviewed since the most recent retraining.

Abstract: A method of forming a keyword based search query that uses a plurality of keywords, in which the keywords of the query are arranged into groups of purpose-related keywords, in which each keyword is associated with a designation of its relative importance. The keywords of a group may be identified manually by the searcher, or existing keywords in a group may be used to suggest or automatically add additional related keywords to the group. The keywords of a group need not be semantically related, but only related to a common purpose of the keyword group. Additional keywords can be suggested to the searcher, or automatically added to the groups. Suitable additional keywords may be identified by reference to previous searches in which the existing keywords were grouped with those additional keywords. Keywords may also be derived from text designated by the searcher, using all or portions of one or more documents or text blocks that the user identifies as describing a concept of interest to the searcher.