Intelligent humanoid interactive content recommender
Methods and systems including computer programs encoded on a computer storage medium, for interactive content recommendation. In one aspect, a method includes receiving a request for content by a user, determining a user intent based on the received request, providing to the user a first attribute responsive to the user intent, receiving a first attribute value responsive to the first attribute, providing a second attribute, and receiving a second attribute value responsive to the second attribute. A particular content vector including a first content attribute and a second content attribute for a particular content item is identified where the first content attribute and the second content attribute sufficiently match the first attribute value and the second attribute value. The particular content item is provided as a suggested content item, and, responsive to a user selection of the particular content item, provided for presentation on the user device.
1. A computer-implemented method for interactive content recommendation, the method being executed by one or more processors and comprising:
generating training data for training a content recommendation model comprising:
receiving a plurality of user feedback items from a plurality of users, each user feedback item including an attribute value responsive to a content attribute describing a particular content item from a plurality of content items;
receiving a plurality of categorical rating variables from the plurality of users, each categorical rating variable is responsive to the particular content item from the plurality of content items;
receiving metadata comprising a plurality of navigational paths for respective selection processes by each user of the plurality of users to select the particular content item of the plurality of content items, wherein each navigational path of the plurality of navigational paths includes attributes provided and attribute values received responsive to the provided attributes during a dialog process interaction; and
generating the training data for the content recommendation model utilizing the plurality of user feedback items, the plurality of categorical rating variables, and metadata for the plurality of content items;
training the content recommendation model utilizing the training data;
receiving, from a user on a user device and by a data processing apparatus, a request for content, the request including a text string input;
determining, by the data processing apparatus and utilizing the content recommendation model, a user intent based on the text string input;
providing, by the data processing apparatus and to the user device, a first attribute, the first attribute responsive to the user intent;
receiving, from the user device and by the data processing apparatus, a first attribute value responsive to the first attribute;
determining, by the data processing apparatus and utilizing the content recommendation model, a second attribute;
providing, by the data processing apparatus and to the user device, the second attribute;
receiving, from the user device and by the data processing apparatus, a second attribute value responsive to the second attribute;
identifying, by the content recommendation model, a particular content vector for a particular content item, wherein the particular content vector includes a first content attribute describing the particular content item and a second content attribute describing the particular content item, and wherein a threshold amount of overlap is determined between the first content attribute and the second content attribute the first attribute value and the second attribute value, respectively;
providing, by the data processing apparatus, the particular content item as a suggested content item to the user on the user device; and
receiving, by the data processing apparatus, a user selection of the particular content item, and in response to the user selection:
providing, for presentation on the user device, the particular content item for presentation on the user device;
providing, to the content recommendation model, user selection data representative of a particular navigational path responsive to the request for content; and
refining the content recommendation model utilizing the user selection data.
2. The method of claim 1 , wherein the content recommendation model further comprises:
a set of themes, each theme comprising a cluster of content items and corresponding user feedback items responsive to the content items,
wherein each theme is assigned a ranking based on the plurality of categorical rating variables for each content item included in the theme.
3. The method of claim 2 , wherein content attribute values describing a particular content item include offline user-provided descriptions of the particular content item.
4. The method of claim 2 , wherein the plurality of categorical rating variables include user preferences regarding different attributes available for content vectors representing the content items.
5. The method of claim 2 , wherein the plurality of user feedback items include structured and unstructured user feedback data.
6. The method of claim 5 , wherein the plurality of user feedback items includes a plurality of social network interactions with a user's social network by a set of other users in the user's social network.
7. The method of claim 6 , wherein user interactions further comprise interactions by the user with the user's social network.
8. The method of claim 2 , wherein a theme of the set of themes comprises a cluster of content items selected based on a mood of the user.
9. The method of claim 1 , wherein receiving a user selection of the particular content item further comprises:
assigning, by the data processing apparatus, a vector attribute score for the particular content vector; and
recording, by the data processing apparatus, the vector attribute score for the particular content vector for the user.
10. One or more non-transitory computer-readable storage media coupled to one or more processors and having instructions stored thereon which, when executed by the one or more processors, cause the one or more processors to perform operations for interactive content recommendation, the operations comprising:
generating training data for training a content recommendation model comprising:
receiving a plurality of user feedback items from a plurality of users, each user feedback item including an attribute value responsive to a content attribute describing a particular content item from a plurality of content items;
receiving a plurality of categorical rating variables from the plurality of users, each categorical rating variable is responsive to the particular content item from the plurality of content items;
receiving metadata comprising a plurality of navigational paths for respective selection processes by each user of the plurality of users to select the particular content item of the plurality of content items, wherein each navigational path of the plurality of navigational paths includes attributes provided and attribute values received responsive to the provided attributes during a dialog process interaction; and
generating the training data for the content recommendation model utilizing the plurality of user feedback items, the plurality of categorical rating variables, and metadata for the plurality of content items;
training the content recommendation model utilizing the training data;
receiving, from a user on a user device and by a data processing apparatus, a request for content, the request including a text string input;
determining, by the data processing apparatus and utilizing the content recommendation model, a user intent based on the text string input;
providing, by the data processing apparatus and to the user device, a first attribute, the first attribute responsive to the user intent;
receiving, from the user device and by the data processing apparatus, a first attribute value responsive to the first attribute;
determining, by the data processing apparatus and utilizing the content recommendation model, a second attribute;
providing, by the data processing apparatus and to the user device, the second attribute;
receiving, from the user device and by the data processing apparatus, a second attribute value responsive to the second attribute;
identifying, by the content recommendation model, a particular content vector for a particular content item, wherein the particular content vector includes a first content attribute describing the particular content item and a second content attribute describing the particular content item, and wherein a threshold amount of overlap is determined between the first content attribute and the second content attribute the first attribute value and the second attribute value, respectively;
providing, by the data processing apparatus, the particular content item as a suggested content item to the user on the user device; and
receiving, by the data processing apparatus, a user selection of the particular content item, and in response to the user selection:
providing, for presentation on the user device, the particular content item for presentation on the user device;
providing, to the content recommendation model, user selection data representative of a particular navigational path responsive to the request for content; and
refining the content recommendation model utilizing the user selection data.
11. The computer storage media of claim 10 , wherein the content recommendation model further comprises:
a set of themes, each theme comprising a cluster of content items and corresponding user feedback items responsive to the content items,
wherein each theme is assigned a ranking based on the plurality of categorical rating variables for each content item included in the theme.
12. The storage media of claim 11 , wherein the plurality of user feedback items include structured and unstructured user feedback items.
13. The storage media of claim 12 , wherein the plurality of user feedback items includes a plurality of social network interactions with a user's social network by a set of other users in the user's social network.
14. The storage media of claim 12 , wherein user interactions further comprise interactions by the user with the user's social network.
15. The storage media of claim 11 , wherein a theme of the set of themes comprises a cluster of content items selected based on a mood of the user.
16. The storage media of claim 10 , wherein receiving a user selection of the particular content item further comprises:
assigning, by the data processing apparatus, a vector attribute score for the particular content vector; and
recording, by the data processing apparatus, the vector attribute score for the particular content vector for the user.
17. A system, comprising:
one or more processors; and
a computer-readable storage device coupled to the one or more processors and having instructions stored thereon which, when executed by the one or more processors, cause the one or more processors to perform operations for interactive content recommendation, the operations comprising:
generating training data for training a content recommendation model comprising:
receiving a plurality of user feedback items from a plurality of users, each user feedback item including an attribute value responsive to a content attribute describing a particular content item from a plurality of content items;
receiving a plurality of categorical rating variables from the plurality of users, each categorical rating variable is responsive to the particular content item from the plurality of content items;
receiving metadata comprising a plurality of navigational paths for respective selection processes by each user of the plurality of users to select the particular content item of the plurality of content items, wherein each navigational path of the plurality of navigational paths includes attributes provided and attribute values received responsive to the provided attributes during a dialog process interaction; and
generating the training data for the content recommendation model utilizing the plurality of user feedback items, the plurality of categorical rating variables, and metadata for the plurality of content items;
training the content recommendation model utilizing the training data;
receiving, from a user on a user device, a request for content, the request including a text string input;
determining, by utilizing the content recommendation model, a user intent based on the text string input;
providing, to the user device, a first attribute, the first attribute responsive to the user intent;
receiving, from the user device, a first attribute value responsive to the first attribute;
determining, utilizing the content recommendation model, a second attribute;
providing, to the user device, the second attribute;
receiving, from the user device, a second attribute value responsive to the second attribute;
identifying, by the content recommendation model, a particular content vector for a particular content item, wherein the particular content vector includes a first content attribute describing the particular content item and a second content attribute describing the particular content item, and wherein a threshold amount of overlap is determined between the first content attribute and the second content attribute the first attribute value and the second attribute value, respectively;
providing the particular content item as a suggested content item to the user on the user device; and
receiving a user selection of the particular content item, and in response to the user selection:
providing, for presentation on the user device, the particular content item for presentation on the user device;
providing, to the content recommendation model, user selection data representative of a particular navigational path responsive to the request for content; and
refining the content recommendation model utilizing the user selection data.
18. The system of claim 17 , wherein the content recommendation model further comprises:
a set of themes, each theme comprising a cluster of content items and corresponding user feedback items responsive to the content items,
wherein each theme is assigned a ranking based on the plurality of categorical rating variables for each content item included in the theme.