System and method for automated bidding using deep neural language models

Patent metadata

Field	Value
Publication number	US-11436628-B2
Application number	US-201715789452-A
Country	US
Kind code	B2
Filing date	Oct 20, 2017
Priority date	Oct 20, 2017
Publication date	Sep 6, 2022
Grant date	Sep 6, 2022

Abstract

Systems, devices, and methods are disclosed for predicting potential effectiveness of query-triggered internet advertisements received from different web page publishers using a deep learning neural network language model for clustering queries, and for automatically adjusting bids for advertisements by advertisers based on the predicted potential effectiveness. Using query-clusters rather than queries for adjusting bids for advertisements allows for more accurate and more consistent bidding strategy despite of sparsity in historical advertisement performance data, higher return on investments for the advertisers, and higher revenue for the publishers of the advertisements.

First claim

What is claimed is:

1. A server, comprising:
a network interface configured to:
receive, from an information-publishing server, an invitation to bid for an advertisement space according to a query received at the information-publishing server from a requester; and
receive, from an advertiser, a bid in response to the query and the invitation to bid; and

a processor in communication with the network interface, configured to:
determine a query cluster, comprising a plurality of search queries, for the query among a predefined set of query clusters using a trained query-clustering neural network, wherein the predefined set of query clusters comprises a first query cluster comprising a first plurality of search queries and a second query cluster comprising a second plurality of search queries;
aggregate historical performance records of query-based advertisements, to obtain aggregated historical performance records, according to the query cluster, comprising the plurality of search queries, as determined by the trained query-clustering neural network;
predict an effectiveness measure of the advertisement space with respect to the query for the advertiser based on the aggregated historical performance records associated with two or more search queries and obtained by the aggregating according to the query cluster, comprising the plurality of search queries, determined for the query associated with the advertisement space; and
automatically control an adjustment to the bid received from the advertiser to obtain an adjusted bid according to the effectiveness measure.

2. The server of claim 1 , wherein the historical performance records of query-based advertisements includes:
a query field specifying queries corresponding to the query-based advertisements;
a device type field specifying types of querying devices corresponding to the query-based advertisements;
a publisher field identifying publishers of the query-based advertisements;
an advertiser field identifying advertisers corresponding to the query-based advertisements; and
a performance indicator field specifying performance indicators corresponding to the query-based advertisements.

3. The server of claim 2 , wherein the performance indicators of the query-based advertisements correspond to conversion rate ratios between rates of conversion of the query-based advertisements and benchmark rates of conversion.

4. The server of claim 3 ,
wherein a conversion corresponds to a querying device browsing a web page associated with a query-based advertisement for a predetermined length of time, the querying device inputting information to the web page, or a purchase by the querying device being recorded on the web page; and
wherein the rates of conversion correspond to numbers of conversion for the query-based advertisements normalized to numbers of browses of the query-based advertisements.

5. The server of claim 2 , wherein the processor is configured to aggregate the historical performance records to obtain the aggregated historical performance records by:
converting the query field to a query cluster field using the trained query-clustering neural network; and
aggregating one or more performance indicators of one or more historical performance records having an identical query cluster field, an identical device type field, an identical publisher field, and an identical advertiser field to obtain the aggregated historical performance records.

6. The server of claim 5 , wherein the processor is configured to predict the effectiveness measure of the advertisement space with respect to the query for the advertiser by:
identifying a record from the aggregated historical performance records having at least one of a query cluster field, a publisher field, an advertiser field, or a device type field that respectively match the query cluster, the information-publishing server, the advertiser, and a device type for the requester; and
predicting the effectiveness measure of the advertisement space with respect to the query for the advertiser based on the one or more performance indicators of the record.

7. The server of claim 2 , wherein the processor is configured to aggregate the historical performance records to obtain the aggregated historical performance records by:
converting the query field to a query cluster field using the trained query-clustering neural network; and
aggregating one or more performance indicators of one or more historical performance records having identical query cluster field, an identical device type field, and an identical publisher field to obtain the aggregated historical performance records.

8. The server of claim 2 , wherein the processor is configured to aggregate the historical performance records to obtain the aggregated historical performance records by:
converting the query field to a query cluster field using the trained query-clustering neural network; and
aggregating one or more performance indicators of one or more historical performance records having an identical query cluster field, an identical device type field, and an identical advertiser field to obtain the aggregated historical performance records.

9. The server of claim 2 , wherein the processor is configured to aggregate the historical performance records to obtain the aggregated historical performance records by:
converting the query field to a query cluster field using the trained query-clustering neural network; and
aggregating one or more performance indicators of one or more historical performance records having an identical query cluster field and an identical device type field to obtain the aggregated historical performance records.

10. The server of claim 1 , wherein the trained query-clustering neural network is configured to convert queries into vector representations.

11. The server of claim 10 , wherein the trained query-clustering neural network is configured to cluster the queries based on clustering the vector representations, wherein the clustering comprises (i) clustering a first query with a second query into the first query cluster based on a similarity between the first query and the second query in the vector representations and (ii) clustering a third query and a fourth query into the second query cluster based on a similarity between the third query and the fourth query in the vector representations.

12. The server of claim 11 , wherein the trained query-clustering neural network is configured to perform the clustering based on a user search query similarity sensitivity level associated with grouping queries into a predetermined number of query clusters.

13. A method, comprising:
receiving, from an information-publishing server, an invitation to bid for an advertisement space according to a query received at the information-publishing server from a requester;
receiving, from an advertiser, a bid in response to the query and the invitation to bid;
determining a query cluster, comprising a plurality of search queries, for the query among a predefined set of query clusters using a trained query-clustering neural network, wherein the predefined set of query clusters comprises a first query cluster comprising a first plurality of search queries and a second query cluster comprising a second plurality of search queries;
aggregating historical performance records of query-based advertisements to obtain aggregated historical performance records according to at least query clusters as determined by the trained query-clustering neural network;
predicting an effectiveness measure of the advertisement space with respect to the query for the advertiser based on the query clu

Assignees

• Yahoo Holdings Inc
• Yahoo Ad Tech Llc

Inventors

• Cetintas Suleyman
• Yang Jian
• Shahshahani Ben

Classifications

• G06N3/09

  Supervised learning · CPC title

• G06N3/0499

  Feedforward networks · CPC title

• G06Q30/02

  Marketing; Price estimation or determination; Fundraising · CPC title

• G06Q30/0273 Primary

  Determination of fees for advertising · CPC title

• G06N3/04

  Architecture, e.g. interconnection topology · CPC title