Content item selection in a digital transaction management platform
A centralized document system identifies content items for presentation to a user based initially on a cold-start algorithm and subsequently based on machine-learned models. The system detects a first access by the user. The system generates a user attribute vector for the user and a content vector for each content item. The system selects a first content item based on the initial cold-start algorithm and modifies a user interface to include the first content item. The system identifies an interaction with the first content item by the user. The system detects a second access by the user. The system selects a machine-learned model based on a set of interactions by the user with displayed content items. The system applies the selected machine-learned model to the set of interactions and the user attribute vector to identify a second content item and modifies the user interface to include the second content item.
1. A method comprising:
detecting, by a centralized document system, a first access by a user to the centralized document system, the user having not previously accessed the centralized document system;
generating, by the centralized document system, a user attribute vector for the user, each user attribute vector entry corresponding to a characteristic of the user;
generating, by the centralized document system, a content vector associated with each of a plurality of content items, each content vector entry corresponding to a characteristic of the associated content item;
selecting, by the centralized document system, a first content item of the plurality of content items based on the user attribute vector and the content vectors;
modifying, by the centralized document system, a user interface displayed to the user to include the first content item;
identifying, by the centralized document system, an interaction by the user with the first content item;
detecting, by the centralized document system, a second access by the user to the centralized document system;
selecting, by the centralized document system, one of a plurality of machine-learned models based on a set of interactions by the user with displayed content items within the centralized document system, the set of interactions including the identified interaction taken by the user with the first content item, each machine-learned model configured to select content items based on user interactions and user characteristics, wherein selecting one of the plurality of machine-learned models comprises:
determining a predictive performance of each of the plurality of machine-learned models based on the set of interactions by the user with displayed content items within the centralized document system; and
selecting a best performing machine-learned model of the plurality of machine-learned models;
applying, by the centralized document system, the selected machine-learned model to the set of interactions and the user attribute vector to identify a second content item of the plurality of content items; and
modifying, by the centralized document system, the user interface displayed to the user to include the second content item.
2. The method of claim 1 , wherein the plurality of machine-learned models includes one or more of: a deep and wide neural network, a deep factorization machine neural network, a graph-based neural network, and a reinforcement learning model.
3. The method of claim 1 , wherein the first access by the user to the centralized document system is a first login to an account of the user after establishing the account with the centralized document system.
4. The method of claim 1 , further comprising:
accessing a training set of historic data comprising a historic user attribute vector, historic content vectors, and historic user interactions;
training the plurality of machine-learned models using the training set of historic data to select content items to present to the user; and
re-training the plurality of machine-learned models using the set of interactions by the user with displayed content items within the centralized document system.
5. The method of claim 1 , wherein selecting the first content item of the plurality of content items based on the user attribute vector and the content vectors comprises:
for each content vector, determining one or more matches between the user attribute vector entries of the user attribute vector and the content vector entries of the content vector;
ranking the plurality of content items based on a number of determined matches; and
selecting a top ranked content item as the first content item.
6. The method of claim 5 , further comprising:
weighting one or more content vector entries of the content vector; and
for any weighted content vector entry without a matching user attribute vector entry, removing the content vector entry from the content vector before ranking the plurality of content items.
7. A non-transitory computer readable storage medium comprising computer executable instructions that when executed by one or more processors causes the one or more processors to perform operations comprising:
detecting, by a centralized document system, a first access by a user to the centralized document system, the user having not previously accessed the centralized document system;
generating, by the centralized document system, a user attribute vector for the user, each user attribute vector entry corresponding to a characteristic of the user;
generating, by the centralized document system, a content vector associated with each of a plurality of content items, each content vector entry corresponding to a characteristic of the associated content item;
selecting, by the centralized document system, a first content item of the plurality of content items based on the user attribute vector and the content vectors;
modifying, by the centralized document system, a user interface displayed to the user to include the first content item;
identifying, by the centralized document system, an interaction by the user with the first content item;
detecting, by the centralized document system, a second access by the user to the centralized document system;
selecting, by the centralized document system, one of a plurality of machine-learned models based on a set of interactions by the user with displayed content items within the centralized document system, the set of interactions including the identified interaction taken by the user with the first content item, each machine-learned model configured to select content items based on user interactions and user characteristics, wherein selecting one of the plurality of machine-learned models comprises:
determining a predictive performance of each of the plurality of machine-learned models based on the set of interactions by the user with displayed content items within the centralized document system; and
selecting a best performing machine-learned model of the plurality of machine-learned models;
applying, by the centralized document system, the selected machine-learned model to the set of interactions and the user attribute vector to identify a second content item of the plurality of content items; and
modifying, by the centralized document system, the user interface displayed to the user to include the second content item.
8. The non-transitory computer readable storage medium of claim 7 , wherein the plurality of machine-learned models includes one or more of: a deep and wide neural network, a deep factorization machine neural network, a graph-based neural network, and a reinforcement learning model.
9. The non-transitory computer readable storage medium of claim 7 , wherein the first access by the user to the centralized document system is a first login to an account of the user after establishing the account with the centralized document system.
10. The non-transitory computer readable storage medium of claim 7 , wherein the instructions further cause the processors to perform operations comprising:
accessing a training set of historic data comprising a historic user attribute vector, historic content vectors, and historic user interactions;
training the plurality of machine-learned models using the training set of historic data to select content items to present to the user; and
re-training the plurality of machine-learned models using the set of interactions by the user with displayed content items within the centralized document system.
11. The non-transitory computer readable storage medium of claim 7 , wherein selecting the first content item of the plurality of content items based on the user attribute vector and the content vectors comprises:
for each content vector, determining one or more matches between the user attribute vector entries of the user attribute vector and the content vector entries of the content vector;
ranking the plurality of content items based on a number of determined matches; and
selecting a top ranked content item as the first content item.
12. The non-transitory computer readable storage medium of claim 11 , wherein the instructions further cause the processors to perform operations comprising:
weighting one or more content vector entries of the content vector; and
for any weighted content vector entry without a matching user attribute vector entry, removing the content vector entry from the content vector before ranking the plurality of content items.
13. A computer system comprising:
one or more computer processors; and
a non-transitory computer readable storage medium comprising computer executable instructions that when executed by one or more processors causes the one or more processors to perform operations comprising:
detecting, by a centralized document system, a first access by a user to the centralized document system, the user having not previously accessed the centralized document system;
generating, by the centralized document system, a user attribute vector for the user, each user attribute vector entry corresponding to a characteristic of the user;
generating, by the centralized document system, a content vector associated with each of a plurality of content items, each content vector entry corresponding to a characteristic of the associated content item;
selecting, by the centralized document system, a first content item of the plurality of content items based on the user attribute vector and the content vectors;
modifying, by the centralized document system, a user interface displayed to the user to include the first content item;
identifying, by the centralized document system, an interaction by the user with the first content item;
detecting, by the centralized document system, a second access by the user to the centralized document system;
selecting, by the centralized document system, one of a plurality of machine-learned models based on a set of interactions by the user with displayed content items within the centralized document system, the set of interactions including the identified interaction taken by the user with the first content item, each machine-learned model configured to select content items based on user interactions and user characteristics, wherein selecting one of the plurality of machine-learned models comprises:
determining a predictive performance of each of the plurality of machine-learned models based on the set of interactions by the user with displayed content items within the centralized document system; and
selecting a best performing machine-learned model of the plurality of machine-learned models;
applying, by the centralized document system, the selected machine-learned model to the set of interactions and the user attribute vector to identify a second content item of the plurality of content items; and
modifying, by the centralized document system, the user interface displayed to the user to include the second content item.
14. The system of claim 13 , wherein the plurality of machine-learned models includes one or more of: a deep and wide neural network, a deep factorization machine neural network, a graph-based neural network, and a reinforcement learning model.
15. The system of claim 13 , wherein the instructions further cause the processors to perform operations comprising:
accessing a training set of historic data comprising a historic user attribute vector, historic content vectors, and historic user interactions;
training the plurality of machine-learned models using the training set of historic data to select content items to present to the user; and
re-training the plurality of machine-learned models using the set of interactions by the user with displayed content items within the centralized document system.
16. The system of claim 13 , wherein selecting the first content item of the plurality of content items based on the user attribute vector and the content vectors comprises:
for each content vector, determining one or more matches between the user attribute vector entries of the user attribute vector and the content vector entries of the content vector;
ranking the plurality of content items based on a number of determined matches; and
selecting a top ranked content item as the first content item.
17. The system of claim 16 , wherein the instructions further cause the processors to perform operations comprising:
weighting one or more content vector entries of the content vector; and
for any weighted content vector entry without a matching user attribute vector entry, removing the content vector entry from the content vector before ranking the plurality of content items.