Abstract: In some examples, systems and methods for generating a hierarchical representation are provided. An example method includes receiving a plurality of purchase orders, generating a plurality of embeddings based on the plurality of purchase orders, clustering the plurality of embeddings into one or more embedding subsets based on a semantic similarity between each embedding of the plurality of embeddings and each other embedding of the plurality of embeddings, and identifying a plurality of labels for the plurality of embedding subsets. In some examples, the method includes generating the hierarchical representation such that the plurality of labels forms a tier in the hierarchical representation. In some examples, subsequent iterations of one or more aspects of the method can be performed to generate additional tiers in the hierarchical representation, where the plurality of labels from a preceding iteration can be used to generate the embeddings for a current iteration.

Abstract: Various embodiments are described herein that relate to communication systems that employ Natural Language Processing (NLP). More specifically, various embodiments relate to systems, methods, and interfaces for performing asynchronous speech act detection on text-based messages transmitted between used of a communication platform. Asynchronous speech act detection allows the content of the messages to be analyzed without interrupting the flow of communication. That is, the messages can be posted for viewing (e.g., to a chat room) and simultaneously transmitted to an NLP server for further analysis. The posted messages can subsequently be updated (e.g., by adding labels that are used for storing, searching, etc.).