International Business Machines Corporation

Kyle Croutwater

Zhe Zhang

Vikrant Verma

Le Zhang

Corville O Allen

Michael D Whitley

Kevin B Haverlock

Dwi Sianto Mansjur

Julius Goth, Iii

Eric Donald Wayne

Beata J Strack

In an approach to generating and using a sentence model for answer generation, one or more computer processors ingest a first corpus of a plurality of text sentences. One or more computer processors convert the plurality of text sentences into a plurality of sentence vectors. One or more computer processors group the plurality of sentence vectors into a plurality of sentence clusters, wherein a sentence cluster is composed of sentences that are semantically similar. One or more computer processors receive a second corpus. One or more computer processors determine, for each sentence cluster of the plurality of sentence clusters, a frequency each sentence cluster appears in the second corpus. Based on the determined frequency, one or more computer processors calculate a probability of each sentence cluster of the plurality of sentence clusters. Based on the calculated probabilities, one or more computer processors generate a first sentence model.

Generating and using a sentence model for answer generation

A method, system, and computer-usable medium are disclosed for identifying areas to improve an interactive conversational system, such as a chatbot. A stream of stream of conversational interactions C (C, C, . . . , C) between users and the interactive conversational system is received. An intent clustering model is periodically applied to the stream to form an incremental clustering based on a set of derived intents to form a mapping from a first set of conversational characteristics to a first set of intents and a second set of conversational characteristics to a first set of unclear intents. Information is provided related to the second set of conversation characteristics.

System and method for alerts for missing coverage of chatbot conversation messages

An approach is provided that receives a question at a question-answering (QA) system. A number of passages are identified that are relevant to the received question. A question knowledge graph is generated that corresponds to the question and a set of passage knowledge graphs are also generated with each passage knowledge graph corresponding to one of the identified passages. Each of the passage knowledge graphs are compared to the question knowledge graph with the comparison resulting in a set of knowledge graph candidate answers (kgCAs). A set of candidate answers (CAs) is computed by the QA with at least one of the CAs being based on one of the kgCAs.

Leveraging entity relations to discover answers using a knowledge graph

Embodiments are directed to a system, a computer program product, and a method for identification of linguistically related elements, and more specifically to prediction of a linguistically related element. A linguistic algorithm forms a cluster representation of corpus entries. A linguistic term is identified and applied to the cluster representation to identify proximally related linguistic terms. Associative relationships between the proximally related terms and category metadata are iteratively investigated. One or more linguistic terms related across the two more metadata categories is identified and designated as the linguistically related element.

Dynamic linguistic assessment and measurement

A computer-implemented method of executing a virtual agent bot includes receiving, via a computer server, at least one input query from a user, and analyzing the at least one input query to extract at least one input term. The method further comprises determining a semantic skill set of the virtual agent bot among a plurality of different candidate skill sets based at least in part on the at least one input term; and invoking the virtual agent bot to provide a semantic topic response corresponding to the semantic skill set to provide an answer to the at least one input query.

Computing system including virtual agent bot providing semantic topic model-based response

Mechanisms are provided in a data processing system for utilizing algorithms based on categories in a question answering system. The mechanisms capture a history of performance and correctness metrics for identifying efficiency of respective algorithms for finding answers to questions in respective question categories in a question answering system. The mechanisms determine sets of algorithms to use for respective question categories according to efficiency and correctness analysis. The mechanisms determine a question category of a given input question and execute a set of algorithms corresponding to the question category of the given input question that meet an efficiency threshold to contribute to finding a correct answer for the given input question.

Learning model for dynamic component utilization in a question answering system

An approach is provided that receives a question at a question answering (QA) system. The question includes a number of words. The approach operates by calculating weights that correspond to search terms included in the plurality of words. The search terms include the plurality of words and may include terms that are one or more sequences of adjacent words included in the question. Based on the calculated weights and the words in the question, the approach generates a query that is used to search a corpus that is managed by the QA system with the search resulting in one or more search results.

Weighting and expanding query terms based on language model favoring surprising words

An active learning framework is operative to identify informative questions that should be added to existing question-answer (Q&A) pairs that comprise a training dataset for a learning model. In this approach, the question-answer pairs (to be labeled as 'true' or 'false') are automatically selected from a larger pool of unlabeled data. A spatial-directed clustering algorithm partitions the relevant question-answer space of unlabeled data. A margin-induced loss function is then used to rank a question. For each question selected, a label is then obtained, preferably by assigning a prediction for each associated question-answer pair using a current model that has been trained on labeled question-answer pairs. After the questions are labeled, an additional re-sampling is performed to assure high quality of the training data. Preferably, and with respect to a particular question, this additional re-sampling is based on a distance measure between correct and incorrect answers.

Geometry-directed active question selection for question answering systems

A computer-implemented method generates a candidate answer triple for use in retrieving information used to answer a question. One or more processors parse a question to identify a lexical answer type for the question, a question action for the question, and a question timestamp for the question to make up a question triple. One or more processors retrieve multiple candidate passages for answering the question, and parse each of the multiple candidate passages to identify a candidate entity, a candidate action, and a candidate timestamp from each of the multiple candidate passages to generate a candidate answer triple. One or more processors compare the question triple to the candidate answer triple and establish a match score for each candidate answer triple, which is used in retrieving information used to answer the question.

Optimizing retrieval of data related to temporal based queries

Mechanisms are provided for configuring a ranker array of a question and answer (QA) system. An initial configuration of a ranker array of the QA system is generated which comprises a plurality of rankers with each ranker applying a statistical classification function to candidate answers generated by the QA system. A training question is input to the QA system to generate a training candidate answer set and the ranker array is applied to the training candidate answer set to generate, for each ranker in the ranker array, a ranked listing of candidate answers. For each ranker in the ranker array, a correctness of a corresponding ranked listing of candidate answers is determined and the ranker array is reconfigured based on results of the determined correctness of the corresponding ranked listing of candidate answers for each ranker in the ranker array.

Growing community of inventors

Chapel Hill, NC, United States of America

Kyle Croutwater