International Business Machines Corporation

Neil Rohit Mallinar

Tin Kam Ho

Rajendra G Ugrani

Ayush Gupta

Abhishek Shah

Luis Alfonso Lastras-Montano

David Amid

David Boaz

Amir Kantor

Robert Leslie Yates

Blake McGregor

A combination of propagation operations and learning algorithms is applied, using a selected set of labeled conversational logs retrieved from a subset of a plurality of conversational logs, to a remaining corpus of the plurality of conversational logs to train an automated response system according to an intent associated with each of the conversational logs. The combination of propagation operations and learning algorithms may include defining the labels by a user for the selected set of the subset of the plurality of conversational logs; training a probabilistic classifier using the defined labels of features of the selected set, wherein the probabilistic classifier produces labeling decisions for the subset of conversational logs; weighting the features of the selected set in a model optimization process; and/or training an additional classifier using the weighted features of the selected set and applying the additional classifier to the remaining corpus.

Intent authoring using weak supervision and co-training for automated response systems

A method, apparatus, and system are provided for resolving conflicts between training data conflicts by retrieving independent training data sets, each comprising a plurality of intents and end-user utterances for use in training one or more classifiers to recognize a corresponding intent from one or more of the end-user utterances, providing a first test end-user utterance associated with a first intent from the first independent training data set to the one or more classifiers to select an output intent generated by the one or more classifiers; identifying a first conflict when the first intent does not match the output intent, and automatically generating, by the system, one or more conflict resolution recommendations for display and selection by an end user to resolve the first conflict.

Method, apparatus, and system for conflict detection and resolution for competing intent classifiers in modular conversation system

Expansion of intent classification data utilizing batch utterance scheduling, by a processor in a computing environment. A set of unlabeled examples for intent processing is received by an intent builder iteratively defining an intent. The set of examples are separated into a first subset processed according to a first model and a second subset processed according to a second model. The first subset is incorporated into the intent builder during a building iteration and scheduling a first batch processing of the second subset processed according to the second model based on a scheduling criteria. The first batch processing of the second subset is initiated once the scheduling criteria is satisfied. Upon completion of the first batch processing, results of the completion are used to influence additional examples retrieved from the first subset and the second subset during a subsequent building iteration by the intent builder.

Iterative intent building utilizing dynamic scheduling of batch utterance expansion methods

Embodiments provide for driving intent expansion via anomaly detection by ranking, according to anomaly scores, a plurality of historic utterances that have been associated by a classifier with a given intent of a plurality of predefined intents; identifying a given utterance from the plurality of historic utterances having a given anomaly score greater than an anomaly threshold; in response to verifying that the given utterance is associated with the given intent, adding the given utterance to a training dataset as a positive example for the given intent; and in response to verifying that the given utterance is not associated with the given intent, adding the given utterance to the training dataset as a complement example for the given intent. A complement example for one intent may be added as a positive example for a different intent. The training dataset may be used to train or retrain an intent classifier.

Driving intent expansion via anomaly detection in a modular conversational system

Evaluating intent authoring processes, by a processor in a computing environment. A dataset comprising utterances of interactive dialog sessions between agents and clients for a given product or service is received. A classification of at least a portion of the utterances is performed for a target intent according to at least one of a plurality of recommendation algorithms, where the classification is performed by an automatic driver invoking the recommendation algorithm and simulating a manual confirmation of the algorithm's decision by a user. A classifier trained with the utterances recommended and confirmed by the automatic driver is automatically evaluated according to at least one of the plurality of evaluation criteria. A report tracking the evaluation results is generated.

Evaluation framework for intent authoring processes

Evaluating intent authoring processes, by a processor in a computing environment. Results are received of a simulated intent labeling effort of a dataset comprising utterances of interactive dialog sessions between agents and clients for a given product or service. Figures of merits for respective algorithms used to perform the simulated intent labeling effort are computed. Each of the respective algorithms are evaluated according to the computed figures of merits; and one of the respective algorithms is implemented for labeling intents of a remaining corpus of the synthesized dataset according to parameters evaluated in the computed figures of merits.

Embodiments for training an automated response system using weak supervision and co-training in a computing environment are provided. A plurality of conversational logs comprising interactive dialog sessions between agents and clients for a given product or service are received. A subset of the plurality of conversational logs are retrieved according to a defined criterion, and a selected set of the subset of the plurality of retrieved conversational logs are labeled by a user. The labeling is associated with a semantic scope of intent considered by the clients. A combination of propagation operations and learning algorithms using the selected set of labeled conversational logs are applied to a remaining corpus of the plurality of conversational logs to train the automated response system according to the semantic scope of intent.

Embodiments for training a conversational system using intent classification are provided. Example utterances recommended to be associated with a particular semantic scope of intent are received from a plurality of sources. Respective ones of the example utterances from the plurality of sources are portioned and stored in a plurality of pools. The respective example utterances from the plurality of pools are collated into a unified pool according to weighting assigned to each one of the plurality of pools, the weighting associated with user actions taken on the respective example utterances. A unified set of the example utterances from the unified pool is output for selection by the user to train an intent classifier associated with the semantic scope of intent for the conversational system.

Intent classification from multiple sources when building a conversational system

A method, system, and apparatus are provided for resolving conflicts between training data conflicts by retrieving independent training data sets, each comprising a plurality of intents and end-user utterances for use in training one or more classifiers to recognize a corresponding intent from one or more of the end-user utterances, providing a first test end-user utterance associated with a first intent from the first independent training data set to the one or more classifiers to select an output intent generated by the one or more classifiers; identifying a first conflict when the first intent does not match the output intent, and automatically generating, by the system, one or more conflict resolution recommendations for display and selection by an end user to resolve the first conflict.

Growing community of inventors

Long Island City, NY, United States of America

Neil Rohit Mallinar