Microsoft Technology Licensing, LLC

Pushpraj Shukla

Mayank Shrivastava

Sonal Prakash Mane

Hui Zhou

Dimitrios Brisimitzis

Emre Hamit Kok

Qifa Ke

Yu Shi

Massimo Mascaro

Andreas Bode

Manish Malik

Rangan Majumder

Jonas Barklund

Hung-An Chang

Atul Kumar Gupta

Yuan Wang

Chhaya Methani

Elliot Kuehl Olds

Sagar Goyal

Dario Vignudelli

Pin-Jung Chen

Sahil Bhatnagar

Ipolitas Clinton Dunaravich

Arko P Mukherjee

The disclosure herein describes a system for generating embeddings representing sequential human activity by self-supervised, deep learning models capable of being utilized by a variety of machine learning prediction models to create predictions and recommendations. An encoder-decoder is provided to create user-specific journeys, including sequenced events, based on human activity data from a plurality of tables, a customer data platform, or other sources. Events are represented by sequential feature vectors. A user-specific embedding representing user activities in relationship to activities of one or more other users is created for each user in a plurality of users. The embeddings are updated in real-time as new activity data is received. The embeddings can be fine-tuned using labeled data to customize the embeddings for a specific predictive model. The embeddings are utilized by predictive models to create product recommendations and predictions, such as customer churn, next steps in a customer journey, etc.

Self-supervised system generating embeddings representing sequenced activity

Training query intents are allocated for multiple training entities into training time intervals in a time series based on a corresponding query intent time for each training query intent. Training performance results for the multiple training entities are allocated into the training time intervals in the time series based on a corresponding performance time of each training performance result. A machine learning model for a training milestone of the time series is trained based on the training query intents allocated to a training time interval prior to the training milestone and the training performance results allocated to a training time interval after the training milestone. Target performance for the target entity for an interval after a target milestone in the time series is predicted by inputting to the trained machine learning model target query intents allocated to the target entity in a target time interval before the target milestone.

Time-factored performance prediction

A computing system including one or more processors generates a topic set for a domain. A taxonomic evaluator is executed by the one or more processors to evaluate a set of category clusters generated from domain-specific textual data against a domain-specific taxonomic tree based on a coherency condition and to identify the category clusters that satisfy the coherency condition. The domain-specific taxonomic tree is generated from hierarchical structures of documents relating to the domain. Each identified category cluster is labeled with a label. A topic set creator is executed by the one or more processors to insert the labels of the set of identified category clusters into the topic set for the domain.

Topic set refinement

Non-limiting examples of the present disclosure describe a unified ranking model that may be used by a plurality of entry points to return ranked results in response to received query data. The unified ranking model is provided as a service for a plurality of entry points. A query is received from an entry point of the plurality of entry points. Results data for the query data is retrieved. A unified ranking model is executed to rank the results data. Execution of the unified ranking model manipulates feature data of the unified ranking model based on user context signals associated with the received query data and acquired result retrieval signals corresponding with the retrieved results data. Execution of the unified ranking model generates ranked result data. Ranked results data is returned to the processing device corresponding with the entry point. Other examples are also described.

Single unified ranker

In various embodiments, systems and methods are provided for query expansion using add-on terms with classifications. A query is received. An add-on term is identified for the query. A classification is determined for the add-on term. The classification is a designation associated with the add-on term that is used to distinguish the add-on term from the query. An appended query is generated based on the add-on term. The appended query is generated by concatenating the query with the add-on term. The appended query is executed on a resource stack as a single reformulated query to identify one or more resources. Upon execution, the classification of the add-on term distinguishes the one or more resources identified for the add-on term based on tagging the one or more resources with the classification of the add-on term. The appended query is used to generate content items.

Query expansion using add-on terms with assigned classifications

Growing community of inventors

Dublin, CA, United States of America

Pushpraj Shukla