Session context modeling for conversational understanding systems

Abstract

Systems and methods are provided for improving language models for speech recognition by adapting knowledge sources utilized by the language models to session contexts. A knowledge source, such as a knowledge graph, is used to capture and model dynamic session context based on user interaction information from usage history, such as session logs, that is mapped to the knowledge source. From sequences of user interactions, higher level intent sequences may be determined and used to form models that anticipate similar intents but with different arguments including arguments that do not necessarily appear in the usage history. In this way, the session context models may be used to determine likely next interactions or "turns" from a user, given a previous turn or turns. Language models corresponding to the likely next turns are then interpolated and provided to improve recognition accuracy of the next turn received from the user.

Description

Background technique [0001] Human-machine dialogue systems that support voice (such as voice interaction with entertainment systems or personal devices) rely on accurate recognition of user voice. For example, an effective voice search application must accurately recognize the query or other interaction submitted by the user so that the information returned to the user is related to the user's intention to submit the query or action. In a series of interactions or "rounds" with one of these systems, users may submit multiple queries. Usually, the content of these queries changes from one round to the next at the word level or vocabulary, but it usually shares some relevance at the semantic or intent level in the same conversation. For example, a user can ask about a movie and then want to know the location near the theater where the movie is being played. [0002] This situation is particularly common in structured domains (such as entertainment systems or personal assistant app...

AI Technical Summary

Problems solved by technology

However, capturing this information at the semantic level with limited past data (e.g., past queries in a particular session up to the current time) is challenging

Furthermore, existing efforts to model session context only consider past queries in the current session and assume that the entire session is only for one specific topic or intent

Additionally, these methods do not model the sequential actions taken by the user in each session

Images