Microsoft Technology Licensing, LLC

Other

Reid Andersen

Kumar Hemachandra Chellapilla

Christian Herwarth Borgs

Jennifer Tour Chayes

John E Hopcroft

Shang-Hua Teng

Chinmay Karande

David Maxwell Chickering

Patrice Y Simard

Kamal Jain

Denis X Charles

Ewa Dominowska

Anton Mityagin

Scott Joseph Counts

Jack W Stokes

David Max Chickering

Sujay Kumar Jauhar

Chirag Deepakkumar Shah

Krishna Chaitanya Gade

Seyed Vahab Mirrokni

Longqi Yang

Yuval Peres

Amit Prakash

Mengting Wan

Sathish Kumar Manivannan

Vahab Mirrokni

Matt Jacobsen

Siddharth Suri

Jennifer Lynay Neville

Georg Ludwig Wilhelm Buscher

Tara Lynn Safavi

Xiaochuan Ni

Various embodiments discussed herein relate to prompting a model, such as a Large Language Model (LLM), to ingest natural language clustering instructions and generate corresponding natural language clustering information, such as a cluster description and/or a cluster label without the need to generate any numeric text embeddings.

Dataset clustering via language model prompts

Systems, methods, and computer storage media having computer-executable instructions embodied thereon that facilitate evaluation of digital content preferences are provided. A user is presented with items of digital content and permitted to manipulate the arrangement of the digital content items in the context of a layout area. Based on the user's manipulation of the digital content items, a user preference regarding an arrangement of digital content, such as a location preference, a position preference, and/or a usage preference, is identified. In embodiments, such a user preference can be utilized to later display digital content to a user in accordance therewith.

Evaluating preferences of content on a webpage

Various embodiments provide techniques for graph clustering. In one or more embodiments, a participation graph is obtained that represents relationships between entities. An auxiliary graph is constructed based on the participation graph. The auxiliary graph may be constructed such that the auxiliary graph is less dense than the participation graph and is therefore computationally less complex to analyze. Clusters in the auxiliary graph are determined by solving an objective function defined for the auxiliary graph. Clusters determined for the auxiliary graph may then be utilized to ascertain clusters in the participation graph that solve a related objective function defined for the participation graph.

Graph clustering

The claimed subject matter relates to an architecture that can identify, store, and/or output local contributions to a rank of a vertex in a directed graph. The architecture can receive a directed graph and a parameter, and examine a local subset of vertices (e.g., local to a given vertex) in order to determine a local supporting set. The local supporting set can include a local set of vertices that each contributes a minimum fraction of the parameter to a rank of the vertex. The local supporting set can be the basis for an estimate of the supporting set and/or rank of the vertex for the entire graph and can be employed as a means for detecting link or web spam as well as other influence-based social network applications.

Local computation of rank contributions

Providing for local graph partitioning using an evolving set process is disclosed herein. By way of example, a computer processor can be configured to execute local partitioning based on evolving set instructions. The instructions can be employed to transition a set of analyzed vertices of a graph until a segment of the graph with small conductance is identified. A transitioning algorithm can expand or contract the analyzed set of vertices based on characteristics of vertices at a boundary of the analyzed set. Accordingly, as the set of analyzed vertices becomes large, significant processing efficiency is gained by employing the characteristics of boundary vertices to transition the set or determine conductance, rather than all vertices of the analyzed set.

Local graph partitioning using evolving sets

Short paths are found with a small query time in scale-free directed graphs using a two-phase process by which data structures comprising shortest path trees are first pre-computed for a group of central vertices called 'hubs' that have short paths to most other vertices in the graph. In a query time phase, a short path between two vertices of interest in the graph is found by looking up the path to the root in each of the shortest path trees.

Short paths in web graphs with small query time

Classes of web graph algorithms are extended to run directly on virtual node-type compressed web graphs where a reduction in runtime of the extended algorithms is realized which is approximately proportional to the compression ratio applied to the original (i.e., uncompressed) graph. In the virtual node compression technique, a succinct representation of a web graph is constructed by replacing dense subgraphs by sparse ones so that the resulting compressed graph has significantly fewer edges and a relatively small number of additional nodes.

Speeding up analysis of compressed web graphs using virtual nodes

An approach for identifying suspect network sites in a network environment entails using one or more malware analysis modules to identify distribution sites that host malicious content and/or benign content. The approach then uses a linking analysis module to identify landing sites that are linked to the distribution sites. These linked sites are identified as suspect sites for further analysis. This analysis can be characterized as 'bottom up' because it is initiated by the detection of potentially problematic distribution sites. The approach can also perform linking analysis to identify a suspect network site based on a number of alternating paths between that network site and a set of distribution sites that are known to host malicious content. The approach can also train a classifier module to predict whether an unknown landing site is a malicious landing site or a benign landing site.

Bottom-up analysis of network sites

The claimed subject matter provides a system and/or a method that facilitates reducing spam in search results. An interface can obtain web graph information that represents a web of pages. A spam detection component can determines one or more features based at least in part on the web graph information. The one or more features can provide indications that a particular page of the web graph is spam. In addition, a robust rank component is provided that limits amount of contribution a single page can provide to the target page.

Locally computable spam detection features and robust pagerank

Growing community of inventors

Seattle, WA, United States of America

Reid Andersen