Back to results
US 11,768,888 B2Granted
Attribute extraction
AssigneePinterest, Inc.
InventorsJinfeng Zhuang, Zhengda Zhao, Vijai Mohan
Life of this patent
Estimating the term…
Abstract
Disclosed are systems and methods for autonomously extracting attributes from domains of a vertical. The disclosed implementations train a deep neural network (“DNN”) based on one or more domains of a vertical using labeled embedding vectors generated for nodes of those one or more domains. The trained DNN may then be used to autonomously label nodes of other domains within the same vertical such that attributes corresponding to those labels can be extracted.
Claims19
A computer-implemented method, comprising:
- for each of a first plurality of nodes of one or more web pages of a first domain of a shopping vertical:
- generating a node embedding vector representative of the node;
- and
- associating a label of a plurality of labels with the node embedding vector to produce labeled data of the first domain, wherein the plurality of labels include at least one of a price label, a title label, an image label, or a description label;
- training a deep neural network (“DNN”) based at least in part on the labeled data of the first domain to produce a trained DNN for the shopping vertical;
- processing a first web page of a second domain of the shopping vertical to determine a second plurality of nodes of the first web page;
- generating a plurality of node embedding vectors, each node embedding vector of the plurality of node embedding vectors representative of a node of the second plurality of nodes;
- determining, based at least in part on the trained DNN and the plurality of node embedding vectors, a label of the plurality of labels for each of the second plurality of nodes;
- extracting from each of the second plurality of nodes and based at least in part on the one or more of the labels, one or more attributes;
- and
- generating, based at least in part on the one or more attributes extracted from the second plurality of nodes, a second web page on a third domain, wherein:
- the second web page is distinct from the first web page;
- the second web page includes at least one of the one or more attributes extracted from the second plurality of nodes;
- the third domain is of a second vertical that is different than the shopping vertical;
- and
- the second web page includes a control button, that when selected, results in at least one of additional information being provided about an item corresponding to the at least one attribute included in the second web page, an ability to purchase the item, or a redirect from the second web page to the first web page.
13 more elements
A computing system, comprising:
- one or more processors;
- and
- a memory storing program instructions that, when executed by the one or more processors, cause the one or more processors to at least:
- for a first web page of a first domain of a vertical:
- parse code of the first web page to determine a plurality of nodes of the first web page;
- determine, for each of the plurality of nodes, a corresponding visual indicator;
- determine a plurality of labels corresponding to the first domain;
- present a labeling interface that includes the plurality of labels and the corresponding visual indicator for each of the plurality of nodes of the first web page;
- receive, in response to an interaction with the labeling interface, an assignment of a label of the plurality of labels to a first node of the plurality of nodes;
- generate a node embedding vector representative of the first node;
- and
- associate the label of the plurality of labels with the node embedding vector to produce labeled data of the first domain;
- train a deep neural network (“DNN”) based at least in part on the labeled data of the first domain to produce a trained DNN for the vertical;
- generate a plurality of node embedding vectors, each node embedding vector of the plurality of node embedding vectors representative of a node of a second plurality of nodes corresponding to a second domain of the vertical, wherein the first domain and the second domain are independent;
- determine, based at least in part on the trained DNN and the plurality of node embedding vectors, a label of the plurality of labels for each of the second plurality of nodes;
- and
- extract, for a second node of the second plurality of nodes, one or more attributes as an attribute corresponding to the label determined for the node.
14 more elements
A method, comprising:
- generating, for a first domain, a plurality of node embedding vectors, each node embedding vector of the plurality of node embedding vectors representative of a node of a plurality of nodes corresponding to the first domain;
- determining, based at least in part on a deep neural network (“DNN”) trained with data that is independent of the first domain and the plurality of node embedding vectors, a label of a plurality of labels for each of the plurality of nodes;
- extracting, for a first node of the plurality of nodes, one or more attributes;
- and
- generating, based at least in part on the one or more attributes extracted for the first node of the plurality of nodes, a second domain that is different than the first domain, wherein:
- the second domain includes at least one of the one or more attributes extracted for the first node;
- the second domain is of a second vertical that is different than a first vertical of the first domain;
- and
- the second domain includes a control button, that when selected, results in at least one of additional information being provided about an item corresponding to the at least one attribute included in the second domain, an ability to purchase the item, or a redirect from the second domain to the first domain.
6 more elements
Dependencies are inferred from the claim text.