Sap Se

Microsoft Technology Licensing, LLC

Sap Se Walldorf

Haichao Wei

Priyanka Khaitan

Kevin Hsu

Mitchell Clark

Huiji Gao

Jun Shi

Qin Iris Wang

Yu Gan

Gonzalo Jorge Aniano Porcile

Yuwei Qiu

Joseph Michael Esposito

Sameer Ulhas Shenai

Pierre Y Monestie

Wenqiong Liu

Anshu Basia

Faezeh Hajiaghajani Memar

Qiang Xiao

Alice Hing-Yee Leung

Lili Zhou

Chih-Hui Wang

Zhiling Wan

A device may provide a prompt to a first machine learning model. The prompt may include at least one instruction to cause the first machine learning model to use at least first natural language input associated with a use of a conversational search system to rank data sources, generate a first search query and reasoning, and use the first search query and the reasoning to generate a second search query. The first search query may include data obtained from at least one of the ranked data sources. The reasoning may include an explanation of how the first machine learning model generated the first search query. A second machine learning model may synthesize a response determined via execution of the second search query. The synthesized response may be provided for presentation via the conversational search system.

Digital content generation with in-prompt hallucination management for conversational agent

In an example, a particular type of deep learning model is used in the global model of the GDMix model: a Factorization Machine. A Factorization Machine combines a Support Vector Machine (SVM) and Matrix Factorizations. It has the advantage of modeling data with huge sparsity well, while maintaining a linear time complexity. A modification may be further made to the Factorization Machine by introducing Lnorm reduction. This acts to divide calculations made by the Factorization Machine into a portion that can be precomputed and a portion that cannot be precomputed. The portion that can be precomputed is then precomputed in an offline manner. As such, when the model is operated in an online manner, the Factorization Machine only needs to compute the portion that cannot be precomputed, reducing the number of operations that need to performed at runtime and greatly improving processing speed over prior machine learned models.

Factorization machine with L-2 norm reduction for machine learned models

Embodiments of the described technologies are capable of reading a text sequence that include at least one word; extracting model input data from the text sequence, where the model input data includes, for each word of the text sequence, segment data and non-segment data; using a first machine learning model and at least one second machine learning model, generating, for each word of the text sequence, a multi-level feature set; outputting, by a third machine learning model, in response to input to the third machine learning model of the multi-level feature set, a tagged version of the text sequence; executing a search based at least in part on the tagged version of the text sequence.

Neural tagger with deep multi-level model

Embodiments of the disclosed technologies use machine learning to produce thread level classification data and case level classification data.

Domain-adapted sentiment prediction for long or unbalanced text threads

Systems and methods are provided for digital assistant configuration and functionality. For example, systems and methods provide for receiving a query from a user via a computing device, processing language in the query to identify a plurality of elements associated with the query, and analyzing the plurality of elements associated with the query to determine an intent of the query by mapping the plurality of elements associated with the query to a list of predetermined intents by comparing the plurality of elements associated with the query to each intent in the list of predetermined intents to generate a score for each intent in the list of predetermined intents. Systems and methods further provide for determining a subset of the predetermined intents based on the score for each intent in the list of predetermined intents, and providing recommendations related to the query based on the subset of predetermined intents.

Digital assistant query intent recommendation generation

In an example, one or more tokens in a freeform search query are identified, each token corresponding to a search term or combination of neighboring search terms in the freeform search query. A graph structure and flat data corresponding to the graph structure are obtained. A search for results relevant to the identified one or more tokens is performed using a first search process, the first search process using the graph structure and flat data to rank search results according to a first ranking order and populate a why found table. Pieces of data corresponding to the results relevant to the identified one or more tokens are joined using the why found table, forming a view and one or more patterns of operators and terms, in a pattern database, matching the one or more operators and one or more surrounding terms in the freeform search query are found.

Internet of things structured query language query formation

Systems and methods are provided for receiving and processing a query to determine an intent of the query, and analyzing the intent of the query to determine that the intent of the query is associated with a use case state diagram. Systems and methods further provide for traversing a plurality of nodes in the user case state diagram to generate a value for each relevant node, performing an operation associated with the query using the generated values, and providing the results of the operation to a computing device.

Dialog system for transitioning between state diagrams

Systems and methods are provided for receiving a plurality of queries from plurality of channels via a plurality of computing devices. For each query of the plurality of queries systems and methods are provided for determining a use case associated with the query from a plurality of predetermined use cases, determining transformation rules for data associated with the use case, accessing data from at least one data source of a plurality of data sources to generate response data for a response to the query, the plurality of data sources comprising the data in a plurality of different data formats, transforming the data associated with the use case from at least a first format into a uniform data structure comprising the response data using the transformation rules for the response data, and providing the response data in the uniform data structure.

Decoupled architecture for query response generation

In an example, data in a non-flat format and metadata corresponding to the data are obtained from a first database. The data is flattened into flat data and augmented with the metadata. One or more pieces of the flat data are scanned to locate a first piece of flat data having a first attribute with attribute values that are a subset of attribute values of a second attribute of a second piece of flat data. A link is then created between the first attribute of the first piece of flat data and the second attribute of the second piece of flat data. A graph structure is generated, the graph structure containing a plurality of nodes, each node corresponding to a data type of the flat data and corresponding to one or more pieces of data in the flat data of the corresponding data type.

Internet of things search and discovery graph engine construction

In an example, a graph structure and flat data corresponding to the graph structure are obtained, the graph structure containing a plurality of nodes. Stop words are removed from one or more search terms to obtain modified search terms. For each of the modified search terms, the pieces of data from the flat data are searched to locate one or more matching pieces of data from the flat data, wherein a piece of data from the flat data matches if it contains at least one attribute with a value that is similar to the modified search term. This information is used to calculate fuzzy match scores for each matching piece of data from the flat data.

Internet of things search and discovery using graph engine

In an example, for each one or more search terms, pieces of data from flat data are searched to locate one or more matching pieces of data from the flat data, wherein a piece of data from the flat data matches if it contains at least one attribute with a value that is similar to the search term. Then, for each matching piece of data from the flat data, a fuzzy match score and a match frequency score are calculated. For each node in a graph structure, a ranking score based on a combination of the fuzzy match score and the match frequency score for a corresponding piece of data from the flat data is calculated. One or more search results are retrurned based on the ranking scores of nodes corresponding to pieces of data for the one or more search results.

Growing community of inventors

Santa Clara, CA, United States of America

Haichao Wei