Safegraph, Inc.

Andrew Israel

Tenzing Shaw

Kaushik Iska

Noah Yonack

Piotr W Kozikowski Kruczkowska

Takashi Yonebayashi

Aaron Goebel

The technology disclosed relates to deduplicating metadata about places. A feature generator module is configured to generate features for metadata profiles. The metadata profiles represent a plurality of places. The features are based on geohash strings and word embeddings generated for the metadata profiles. A diff generator module is configured to generate diff vectors that pair-wise encode results of comparison between features of paired metadata profiles. A classification module is configured to generate similarity scores for the paired metadata profiles based on the diff vectors. A particular similarity score indicates whether metadata profiles in a particular pair of metadata profiles represent a same place.

Feature-based deduplication of metadata for places

The technology disclosed relates to determining visitor behavior. In particular, it relates to receiving preliminary geospatial coordinates of a plurality of places-of-interest, generating (i) contours of the places-of-interest and (ii) hyper-located geospatial coordinates of centroids of the places-of-interest based on processing the preliminary geospatial coordinates of the places-of-interest, and correlating (i) the contours and (ii) the hyper-located geospatial coordinates of the centroids of the places-of-interest with geospatial coordinates of visitor locations to determine the visitor behavior of visitors visiting the places-of-interest, including at least one of detecting whether the visitors visited the places-of-interest, and detecting a duration for which the visitors visited the places-of-interest.

Visitor insights based on hyper-locating places-of-interest

The technology disclosed uses machine learning-based models to compare and merge metadata profiles from various metadata sources to create a library of metadata profiles with the most accurate attributes. The technology disclosed adjusts the geospatial coordinates of places located inside a building structure for accuracy and defines a block for each place within the perimeter of the building structure. The building structure may host a single place or multiple places. The non-overlapping blocks, the adjusted geospatial coordinates of places and more accurate metadata profiles enable the gathering of accurate location-specific data from visitors inside the building structures through their personal devices.

Deduplication of metadata for places

The technology disclosed adjusts the geospatial coordinates of places-of-interest located inside a building structure for accuracy and defines a block for each place-of-interest within the perimeter of the building structure. The building structure may host a single place-of-interest or multiple places-of-interest. The technology disclosed outlines a contour of the building structure based on geospatial coordinates of the perimeter points of the building structure, identifies contained places-of-interest within the contour, divides the contour into pieces, calculating splines for each of the pieces, and partitions the pieces into non-overlapping blocks by orthogonally projecting block boundaries onto the splines. Each contained places-of-interest within the building structure is attributed to a block. The non-overlapping blocks and adjusted geospatial coordinates of places-of-interest enable the gathering of accurate location-specific data from visitors inside the building structures through their personal devices.

Growing community of inventors

San Francisco, CA, United States of America

Andrew Israel