Abstract: A method, system, and computer program product obtain models pretrained on a set of features and map a set of policies to the models. A request to remove at least one feature from the features and, in response to the receiving, a policy from the set of policies that is affected by the at least one feature is identified. The identifying uses an association rule mining (ARM) model. Using a performance evaluation model, performance scores for the models with and without the at least one feature are generated. A reward function is calculated for the at least one feature based on the identifying and the performance scores. When it is determined that the reward function is greater than a threshold value, a recommendation is generated for a user.

Abstract: A method, computer system, and a computer program product for automatically generated question suggestions is provided. The present invention may include generating an intent-entity sequence representative of a current chat conversation with a user. The present invention may also include predicting, using a trained intent-entity prediction model, a next intent-entity sequence based on the generated intent-entity sequence representative of the current chat conversation. The present invention may further include converting the predicted next intent-entity sequence into at least one likely question suggestion to return back to the user.

Abstract: Generating questions by receiving user utterance data, determining an intent confidence vector for the user utterance data, predicting, by a trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector, and generating a next question using the next user-intent confidence vector.

Abstract: A method, system, and computer program product for routing queries to answer resources based on component parts and intents of a received query is provided. The method receives a query from a user. The query is analyzed to identify a set of entities associated with the query and generate an utterance representing the query. The method generates an intent classification for the utterance and a vector for the query. The vector is generated based on the set of entities, the utterance, and the intent classification. The method determines an answer resource for the query based on the vector and the intent classification of the query. In response to determining the answer resource, the method provides an answer interface based on the query, the vector, and the intent classification. The answer interface dynamically provides a response to the query.

Abstract: A computer-implemented method, a computer system, and a computer program product for automatically generated conversation output is provided. The present invention may include training an answer-intent classifier to associate an intent expressed in an example question with an example answer that is responsive to the example question. The present invention may further include classifying, using the trained answer-intent classifier, a content transmitted to the trained answer-intent classifier with the intent expressed in the example question in response to determining, using the trained answer-intent classifier, that the content includes relevant information for generating the example answer that is responsive to the example question.

Abstract: A method, computer system, and a computer program product for automatically generated question suggestions is provided. The present invention may include generating an intent-entity sequence representative of a current chat conversation with a user. The present invention may also include predicting, using a trained intent-entity prediction model, a next intent-entity sequence based on the generated intent-entity sequence representative of the current chat conversation. The present invention may further include converting the predicted next intent-entity sequence into at least one likely question suggestion to return back to the user.

Abstract: An approach is disclosed for storing information tailored to a user and a condition to be accessed offline to fit on a local device. The information is retrieved from a set of online sources by a device. The retrieved information is tailored according to a criteria and the condition to form a curated data. An index is provided to access the curated data. The index is used to answer queries. Responsive to identifying a question without an answer found in the curated data when the device is off-line, searching for the answer when the device is online. Responsive to finding the answer online, pruning the curated data by eliminating lower importance data as needed to store higher importance data based on an availability of space allocated to store the curated data on the local storage device, and updating the curated data stored on the local device.

Abstract: In an approach to customizing meeting notes, a computer receives audio input of a virtual meeting, converts the audio input to text, and displays the text to a plurality of meeting participants. A computer receives highlighted phrases of the text from the plurality of meeting participants and determines a highlighting frequency of each of the highlighted phrases. A computer determines phrases with a highlighting frequency greater than a pre-defined threshold. A computer orders the phrases based on a chronological order of the phrases in the audio input. A computer determines preferences of a first meeting participant associated with a meeting summary. A computer generates a customized summary of the virtual meeting for the first meeting participant of the plurality of meeting participants based on the ordered phrases with a high frequency of highlighting and on the preferences. A computer transmits the customized summary to the first meeting participant.

Abstract: Generating questions by receiving user utterance data, determining an intent confidence vector for the user utterance data, predicting, by a trained next user-intent prediction model, a next user-intent confidence vector using the intent confidence vector, and generating a next question using the next user-intent confidence vector.

Abstract: In an approach for identifying when transferring a real-time conversation on a chatbot application to a customer support agent would be more valuable than outputting suggested queries, a processor classifies a query inputted into a real-time conversation as an intention. A processor predicts a set of next intentions of the user. A processor filters out one or more intentions from the set of next intentions of the user that do not further the real-time conversation. A processor generates a set of suggested queries the user can ask to further the real-time conversation from the subset of next intentions. A processor builds a suggestion evaluation contextual bandit model that determines whether the real-time conversation will be resolved successfully if transferred to a customer support agent. A processor trains the suggestion evaluation contextual bandit model against a set of annotated historical interactions. A processor outputs a response recommendation.

Abstract: When a chatbot application is in a mode of live chat conversation with a user, a trained intent classifier determines an intent that underlies a first live question received by the trained intent classifier from the user. A trained next predictor receives the intent from the intent classifier. The trained next intent predictor generates a set of predicted next intents responsive to receiving the intent. A trained re-ranker selects at least one of the predicted next intents responsive to receiving the set of predicted next intents. A question selection engine sends at least one suggested question to the user responsive to receiving the at least one predicted next intent. As a result of the above, the chatbot application provides to the user at least one suggested next question the user may wish to ask in response to the first question from the user.

Abstract: In an approach to customizing meeting notes, a computer receives audio input of a virtual meeting, converts the audio input to text, and displays the text to a plurality of meeting participants. A computer receives highlighted phrases of the text from the plurality of meeting participants and determines a highlighting frequency of each of the highlighted phrases. A computer determines phrases with a highlighting frequency greater than a pre-defined threshold. A computer orders the phrases based on a chronological order of the phrases in the audio input. A computer determines preferences of a first meeting participant associated with a meeting summary. A computer generates a customized summary of the virtual meeting for the first meeting participant of the plurality of meeting participants based on the ordered phrases with a high frequency of highlighting and on the preferences. A computer transmits the customized summary to the first meeting participant.

Abstract: A method, system, and computer program product for routing queries to answer resources based on component parts and intents of a received query is provided. The method receives a query from a user. The query is analyzed to identify a set of entities associated with the query and generate an utterance representing the query. The method generates an intent classification for the utterance and a vector for the query. The vector is generated based on the set of entities, the utterance, and the intent classification. The method determines an answer resource for the query based on the vector and the intent classification of the query. In response to determining the answer resource, the method provides an answer interface based on the query, the vector, and the intent classification. The answer interface dynamically provides a response to the query.

Abstract: An approach is disclosed for a method, a system, and a computer program product for storing information tailored to a user and a condition to be accessed offline on a local device. The information is retrieved from a set of online sources. The retrieved information is tailored according to a criteria and the condition to form a curated data. The curated data is on the local device. An index is provided to access a selected portion of the curated data. The index is used to answer queries from the user. Responsive to identifying a question without an answer identified in the retrieved content, searching for the answer when the device is online. Responsive to finding the answer online, updating the curated data stored on the local device.

Abstract: For a chat bot, first intent data is provided, including a first instance of sample utterances in a first natural language and intents corresponding to the respective sample utterances. The sample utterances are translated by a computer system to a second natural language as a second instance of the sample utterances. The sample utterances of the second instance are translated by a computer system back to the first natural language as a third instance of the sample utterances. Second intent data is provided including the third instance of the sample utterances and the corresponding intents. An intent classifier of the chat bot is trained to identify respective intents of real time utterances, wherein the identifying is responsive to the intent classifier receiving the real time utterances from a user when the chat bot is in an operating mode and the training includes training the intent classifier on the first and second intent data.

Abstract: For a chat bot, first intent data is provided, including a first instance of sample utterances in a first natural language and intents corresponding to the respective sample utterances. The sample utterances are translated by a computer system to a second natural language as a second instance of the sample utterances. The sample utterances of the second instance are translated by a computer system back to the first natural language as a third instance of the sample utterances. Second intent data is provided including the third instance of the sample utterances and the corresponding intents. An intent classifier of the chat bot is trained to identify respective intents of real time utterances, wherein the identifying is responsive to the intent classifier receiving the real time utterances from a user when the chat bot is in an operating mode and the training includes training the intent classifier on the first and second intent data.

Abstract: A computer-implemented method, a computer system, and a computer program product for automatically generated conversation output is provided. The present invention may include training an answer-intent classifier to associate an intent expressed in an example question with an example answer that is responsive to the example question. The present invention may further include classifying, using the trained answer-intent classifier, a content transmitted to the trained answer-intent classifier with the intent expressed in the example question in response to determining, using the trained answer-intent classifier, that the content includes relevant information for generating the example answer that is responsive to the example question.

Abstract: When a chatbot application is in a mode of live chat conversation with a user, a trained intent classifier determines an intent that underlies a first live question received by the trained intent classifier from the user. A trained next predictor receives the intent from the intent classifier. The trained next intent predictor generates a set of predicted next intents responsive to receiving the intent. A trained re-ranker selects at least one of the predicted next intents responsive to receiving the set of predicted next intents. A question selection engine sends at least one suggested question to the user responsive to receiving the at least one predicted next intent. As a result of the above, the chatbot application provides to the user at least one suggested next question the user may wish to ask in response to the first question from the user.