International Business Machines Corporation

Kristen Maria Summers

Charles Evan Beller

Edward Graham Katz

Palani Sakthi

Christopher F Ackermann

William G Dubyak

Stephen Arthur Boxwell

Michael Drzewucki

Christopher Phipps

Julie T Yu

Rob High

Andrew Doyle

Brian L Keith

Joshua G Hong

Robert A Sheets

David McQuenney

Stanley John Vernier

Keith Gregory Frost

Kyle Matthew Brake

Mohamed Nooman Ahmed

John A Riendeau

John J Thomas

Sean Thomas Thatcher

Andeep S Toor

David McQueeney

An embodiment includes tokenizing an input passage into an n-gram sequence of tokens. The embodiment also includes evaluating the input passage using a trained classification model that generates an output indicative of a classification of the input passage. The embodiment also includes generating a first token vector for a first token of the sequence of tokens and projecting the first token vector to a higher dimensional space, resulting in a first projected token vector. The embodiment also includes generating a first similarity score for the first projected token vector based on comparisons of the first projected token vector to each of a plurality of class representations. The embodiment also includes generating a ranked list of the tokens, wherein the generating of the ranked list includes ranking the first token among others of the tokens based on the first similarity score.

Explainable passage classification

Mechanisms are provided to minimize personally identifiable information (PII) in an electronic document. An iterative personally identifiable information minimization (IPIIM) engine receives an electronic document comprising natural language content having a mention of a protected entity and obfuscates the mention of the protected entity to thereby generate a minimized natural language content. A question answering system processes the minimized natural language content to generate a listing of candidate answers and corresponding confidence scores and the IPIIM engine determines whether or not the minimized natural language content is sufficiently obfuscated based on the listing of candidate answers and corresponding confidence scores. In response to determining that the minimized natural language content is sufficiently obfuscated, the minimized natural language content is provided for processing by a requestor computing device.

Cognitive iterative minimization of personally identifiable information in electronic documents

A computer device receives a request to search a corpus of documents for an entity, wherein the request includes a non-name identifier of the entity. The computing device identifies entries of text within the corpus of documents that reference the non-name identifier. The computing device applies natural language processing (NLP) to content associated with the identified entries within the corpus of documents, wherein the NLP identifies candidate entities associated with the non-name identifier. The computing device selects an entity from the candidate entities based, at least in part, on distances between the candidate entities and references to the non-name identifier in the identified entries. The computing device returns the selected entity to a submitter of the request.

Entity search system

An embodiment for expanding a search query is provided. The embodiment may include receiving a stopping criterion for stopping a search. The embodiment may also include receiving an initial search query. The embodiment may further include submitting the initial search query to an information retrieval system. The embodiment may also include identifying enrichment terms from the retrieved initial set of documents. The embodiment may further include generating a subsequent search query that includes one or more enrichment terms from the retrieved initial set of documents. The embodiment may also include submitting the subsequent search query to the information retrieval system. The embodiment may further include determining whether the stopping criterion is met, and in response to determining the stopping criterion is not met, iterating identifying, generating, submitting steps until the stopping criterion is met.

Iterative query expansion for document discovery

A method learns approximate translations of unfamiliar measurement units during deep question answering (DeepQA) system training and usage. The DeepQA system receives a training set containing Question-Answer (QA) pairs having known unit-of-measurement terms, where each QA pair contains an answer having a known numeric value for a corresponding question from the QA pair. The DeepQA system receives a question from each QA pair from the training set to the DeepQA system in order to find answers and passage phrases to the question from each QA pair, and then identifies all found answers and passage phrases having values that are within a predetermined range of answer values of the training set, where one or more of the identified all found answers and passage phrases contain unfamiliar unit-of-measurement terms, in order to learn approximate translations of the unfamiliar unit-of-measurement terms.

Learning approximate translations of unfamiliar measurement units during deep question answering system training and usage

A candidate document is received, for example, by a document filter. A determination is made based on the content of the candidate document, whether the candidate document is relevant to a document corpus. A determination is made based on the content of the candidate document, whether the candidate document is novel with respect to the document corpus. In response to determining that the candidate document is relevant to the document corpus and novel with respect to the document corpus, the candidate document is added to the document corpus to make at least a portion of the content of the candidate document available for a response to a search query.

Automated determination of document utility for a document corpus

Embodiments relate to an intelligent computer platform to selectively amend one or more document elements. A first document is subjected to natural language processing (NLP) and two or more document characteristics are subjected to an assessment to produce a characteristic value. The document characteristics and corresponding characteristic values are analyzed to produce a characteristic profile for each identified document characteristic. Upon receipt of a new document, document characteristic data and corresponding characteristic value(s) are identified. The corresponding characteristic value(s) of the new document is applied against the produced characteristic profile. New document characteristic data is selectively amended responsive to the comparison, and a new document version is created from the selective amendment.

Removal of personality signatures

A method utilizes a deep question answering (QA) system to provide an answer, to a certain type of question, that includes an unfamiliar measurement unit. An answer key is utilized to train a DeepQA system to search for passages that answer a certain type of question, where the DeepQA system outputs an answer key value and an answer key measurement unit that is associated with the answer key value. The method identifies a candidate answer that includes a candidate passage containing the answer key value but not the answer key measurement unit, where a candidate passage measurement unit in the candidate passage is associated with the answer key value. The method then matches the answer key measurement unit to the candidate passage measurement unit based on the answer key measurement unit and the candidate passage measurement unit both being associated with the answer key value.

Learning unfamiliar measurement units in a deep question answering system

A system and a method for determining user synonyms in a query processing system is disclosed. A first query is received from a user of the query processing system. The query is processed through the query processing system to generate results for the first query. The user then provides a second query. The system determines a contextual similarity between the first query and the second query. When the first query and the second query are determined to be contextually similar, a first term is identified in the first query that is different from a second term in the second query. Once identified, the system determines if the first term and the second term are not listed as synonyms in the thesaurus. If they are not listed as synonyms they can be added as synonyms to the thesaurus.

Learning user synonyms from sequenced query sessions

Embodiments are directed to interaction with an open-domain question and answer system by recognizing questions that are highly broad or abstract, and generating and processing a batch of questions expressing alternate, concrete instances of the more general, abstract question. Responses to all of the questions in the batch are considered as candidates, and the strongest general answers are returned. A weighted, ranked answer set, based on weighting individual concrete questions and scaling the answers proportional to the weight of the questions, is provided to the user. The approach, according to embodiments herein, addresses the need for responses to broad questions in which a response to any of a set of more concrete question instances may serve to answer the question.

Dispersed batch interaction with a question answering system

Embodiments relate to an intelligent computer platform to selectively amend one or more tokens in a document. A first document set is subjected to natural language processing (NLP) and a vector score is identified for two or more documents of the first document set. Upon receipt of a new document, the new document is subjected to NLP and a new document vector score is identified. The new document is analyzed against the first document set, and the identified vector score of the first document set is compared to the vector score of the new document. One or more tokens of the new document are amended responsive to the comparison, and a new document version is created from the selective amendment.

Document anonymization including selective token modification

Embodiments for identifying entities relevant to queries are provided. At least one query is received from a user. The at least one query is associated with at least one entity. Results of the at least one query are analyzed to identify related entities. The related entities are analyzed based on a relevancy score and an information enhancement score for each of the related entities to generate a ranking of the related entities. At least one of the related entities is provided to the user based on the ranking of the related entities.

Methods and systems for providing suggestions to complete query sessions

The present invention may receive the question, a plurality of the candidate answers, and a plurality of documents associated with the plurality of candidate answers in the natural language. Then the present invention may tokenize the question, the plurality of the candidate answers, and the plurality of the documents into a corresponding n-gram sequence. The present invention may map n-gram elements from the tokenized question to the n-gram elements of the plurality of the tokenized candidate answers and the plurality of the tokenized documents using the latent token representation technique. The present invention may score the plurality of tokenized candidate answers based on the latent token representation technique. Then, the present invention may determine the precise answer based on the plurality of scored candidate answers.

Latent token representations for passage and answer scoring in question answering systems

A service, in response to receiving a question in a natural language format, identifies one or more selected passages from a corpus that are relevant to a focus of the question from among multiple passages in the corpus. The service aligns one or more answer grammatical properties of one or more answers, selected from the one or more selected passages, to one or more question grammatical properties of the focus of the question. The service returns the one or more answers in response to the question.

Efficient grammatical property alignment for a question answering system

Batch interaction with a computerized question answering system can produce an answer that more closely relates to a user's information need. A batch of questions can be generated interactively, and provides a context for a first question received from a user. The batch of questions includes or more additional questions which have terms with a nonsynonymous semantic relation to a first term in the first question. A question answering system can process the batch of questions to determine candidate answers. An answer to the first question can be determined based, at least in part, on a combined ranking of the candidate answers.

Growing community of inventors

Takoma Park, MD, United States of America

Kristen Maria Summers