Utilizing a touchpoint attribution attention neural network to identify significant touchpoints and measure touchpoint contribution in multichannel, multi-touch digital content campaigns

Patent metadata

Field	Value
Publication number	US-11287894-B2
Application number	US-201815917052-A
Country	US
Kind code	B2
Filing date	Mar 9, 2018
Priority date	Mar 9, 2018
Publication date	Mar 29, 2022
Grant date	Mar 29, 2022

Abstract

The present disclosure relates to systems, non-transitory computer-readable media, and methods for generating and utilizing a touchpoint attribution attention neural network to identify and measure performance of touchpoints in digital content campaigns. For example, a deep learning attribution system trains a touchpoint attribution attention neural network using touchpoint sequences, which include user interactions with content via one or more digital media channels. In one or more embodiments, the deep learning attribution system utilizes the trained touchpoint attribution attention neural network to determine touchpoint attributions of touchpoints in a target touchpoint sequence. In addition, the deep learning attribution system can utilize the trained touchpoint attribution attention neural network to generate conversion predictions for target touchpoint sequences and to provide targeted digital content over specific digital media channels to client devices of individual users.

First claim

What is claimed is:

1. In a digital environment for distributing electronic content across computing devices utilizing a plurality of digital media channels, a computer-implemented method for utilizing machine-learned attribution models to generate accurate attribution for individual touchpoints in digital content campaigns, the computer-implemented method comprising:
generating a digital target touchpoint path of a target user, the digital target touchpoint path comprising a digital target touchpoint sequence of digital touchpoints and a positive conversion indicator; and
based on identifying the positive conversion indicator, generating target attention weights for the digital touchpoints in the digital target touchpoint sequence utilizing a touchpoint attribution attention neural network by:
generating an encoded touchpoint vector encoding the digital target touchpoint path via an encoding layer of the touchpoint attribution attention neural network;
determining a hidden state vector via a long short-term memory (LSTM) layer of the touchpoint attribution attention neural network from the encoded touchpoint vector, wherein the hidden state vector comprises historical contextual information of the digital target touchpoint sequence; and
generating the target attention weights from the hidden state vector via a touchpoint attention layer of the touchpoint attribution attention neural network that determines attention weights for a touch point sequence based on touchpoints and an order of the touchpoints in the touch point sequence, wherein the target attention weights comprise attention coefficient values that indicate attribution levels for each of the digital touchpoints in the digital target touchpoint sequence.

2. The computer-implemented method of claim 1 , further comprising:
determining a time-decay parameter for a first digital touchpoint of the digital target touchpoint path, wherein the time-decay parameter is based on an elapsed time between a first time of the first digital touchpoint and an end time of the digital target touchpoint path; and
generating an attention weight for the first digital touchpoint utilizing the touchpoint attention layer within the touchpoint attribution attention neural network by applying a time-decayed attention weight for the first digital touchpoint based on the time-decay parameter.

3. The computer-implemented method of claim 1 , further comprising:
jointly training a user bias control machine-learning model and the touchpoint attention layer within the touchpoint attribution attention neural network; and
generating the target attention weights of the digital touchpoints in the digital target touchpoint path utilizing the touchpoint attention layer jointly trained with the user bias control machine-learning model.

4. The computer-implemented method of claim 1 , further comprising utilizing a touchpoint context vector and hidden state vectors within the touchpoint attention layer of the touchpoint attribution attention neural network to generate the target attention weights for the digital target touchpoint sequence, wherein the touchpoint context vector is a high-level representation of a fixed sequence of digital touchpoints that biases one or more digital touchpoints in the digital target touchpoint sequence; and
wherein the target attention weights for each of digital touchpoints the digital target touchpoint sequence are relative to the other digital touchpoints in the digital target touchpoint sequence.

5. The computer-implemented method of claim 4 , further comprising training the touchpoint attribution attention neural network by:
utilizing a plurality of digital training touchpoint paths to generate attention weights of digital touchpoints in digital training touchpoint sequences corresponding to the plurality of digital training touchpoint paths;
generating a digital touchpoint sequence representation of a digital training touchpoint sequence by combining attention weights of digital touchpoints in the digital training touchpoint sequence weighted with corresponding hidden state vectors; and
tuning the touchpoint attention layer based on comparing the digital touchpoint sequence representation with a corresponding digital training conversion.

6. A non-transitory computer-readable medium storing instructions that, when executed by at least one processor, cause a computer system to:
identify a digital target touchpoint sequence of a target user;
generate a plurality of digital target touchpoint sequences by:
appending a first potential touchpoint to the digital target touchpoint sequence to generate a first digital target touchpoint sequence; and
appending a second potential touchpoint to the digital target touchpoint sequence to generate a second digital target touchpoint sequence;

generate target attention weights for digital touchpoints in the plurality of digital target touchpoint sequences utilizing a touchpoint attribution attention neural network by:
generating encoded touchpoint vectors by encoding the digital target touchpoint sequences via an encoding layer of the touchpoint attribution attention neural network;
determining hidden state vectors via a long short-term memory (LSTM) layer of the touchpoint attribution attention neural network from the encoded touchpoint vectors, wherein the hidden state vectors comprise historical contextual information of the digital target touchpoint sequences; and
generating the target attention weights from the hidden state vectors via a touchpoint attention layer of the touchpoint attribution attention neural network that determines attention weights for touch point sequences based on touchpoints and an order of the touchpoints in the touch point sequences, wherein the target attention weights comprise attention coefficient values that indicate attribution levels for each of the digital touchpoints in the digital target touchpoint sequences;

identify digital conversion prediction scores from the target attention weights of the digital touchpoints in the plurality of digital target touchpoint sequences utilizing the touchpoint attribution attention neural network;
select the first potential touchpoint for distributing digital content based on determining that a first conversion prediction score for the first digital target touchpoint sequence indicates a greater likelihood of conversion than and a second conversion prediction score for the second digital target touchpoint sequence.

7. The computer-readable medium of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the computer system to generate a first attention weight for a first digital touchpoint in the first digital target touchpoint sequence by determining a time-decay parameter for the first digital touchpoint, wherein the time-decay parameter is based on an elapsed time between a first time of the first digital touchpoint and an end time of the first digital target touchpoint sequence.

8. The computer-readable medium of claim 6 , wherein the touchpoint attribution attention neural network comprises a user bias control machine-learning model; and
further comprising instructions that, when executed by the at least one processor, cause the computer system to jointly train the user bias control machine-learning model with the encoding layer, the LSTM layer, and the touchpoint attention layer.

9. The computer-readable medium of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the computer system to:
identify a positive conversion indication associated with the first digital target touchpoint sequence; and
provide, for display to a graphical user interface of an admin

Assignees

• Adobe Inc

Inventors

• Yan Zhenyu
• Kumar Fnu Arava Venkata Kesava Sai
• Dong Chen
• Pani Abhishek
• Li Ning

Classifications

• G06N3/048

  Activation functions · CPC title

• G06N3/044

  Recurrent networks, e.g. Hopfield networks · CPC title

• G06N3/084 Primary

  Backpropagation, e.g. using gradient descent · CPC title

• G06N3/0442

  characterised by memory or gating, e.g. long short-term memory [LSTM] or gated recurrent units [GRU] · CPC title

• G06N3/09

  Supervised learning · CPC title