ADAPTIVE PAIRWISE PREFERENCES IN RECOMMENDERS
Methods, systems, and products adapt recommender systems with pairwise feedback. A pairwise question is posed to a user. A response is received that selects a preference for a pair of items in the pairwise question. A latent factor model is adapted to incorporate the response, and an item is recommended to the user based on the response.
1 . (canceled)
2 . A method, comprising:
sending a sequence of questions from a server to a client device, each question in the sequence of questions soliciting one content item from a pair of different content items;
receiving at the server successive responses from the client device to the sequence of questions, each successive response selecting a preference for the one content item in the pair of different content items in each question;
sending the successive responses as feedback to a latent factor model for recommending content to the client device; and
receiving, after each successive response, a probability of a preference for another content item in another pair of different content items.
3 . The method of claim 2 , comprising incorporating adaptive pairwise preference feedback into the latent factor model, such that previous feedback on pairs of different content items affects future pairs of different content items.
4 . The method of claim 3 , wherein the latent factor model operates in a Bayesian framework.
5 . The method of claim 2 , comprising generating a recommendation for content based on the successive responses.
6 . The method of claim 2 , comprising querying for one question in the sequence of questions.
7 . The method of claim 2 , comprising predicting a difference in the preference for the pair of different content items.
8 . The method of claim 2 , comprising determining a change in entropy after each successive response.
9 . A system, comprising:
a processor; and
memory storing code that when executed causes the processor to perform operations, the operations comprising:
sending a sequence of questions from a server to a client device, each question in the sequence of questions soliciting one content item from a pair of different content items;
receiving at the server successive responses from the client device to the sequence of questions, each successive response selecting a preference for the one content item in the pair of different content items in each question;
sending the successive responses as feedback to a latent factor model for recommending content to the client device; and
receiving, after each successive response, a probability of a preference for another content item in another pair of different content items.
10 . The system of claim 9 , comprising incorporating adaptive pairwise preference feedback into the latent factor model, such that previous feedback on the pairs of different content items affects future pairs of different content items.
11 . The system of claim 10 , wherein the latent factor model operates in a Bayesian framework.
12 . The system of claim 9 , wherein the operations comprise generating a recommendation for content based on the successive responses.
13 . The system of claim 9 , wherein the operations comprise querying for one question in the sequence of questions.
14 . The system of claim 9 , wherein the operations comprise predicting a difference in the preference for the pair of different content items.
15 . The system of claim 9 , wherein the operations comprise determining a change in entropy after each successive response.
16 . A memory storing instructions that when executed cause a processor to perform operations, the operations comprising:
sending a sequence of questions from a server to a client device, each question in the sequence of questions soliciting one content item from a pair of different content items;
receiving at the server successive responses from the client device to the sequence of questions, each successive response selecting a preference for the one content item in the pair of different content items in each question;
sending the successive responses as feedback to a latent factor model for recommending content to a user of the client device; and
receiving, after each successive response, a probability of a preference for another content item in another pair of different content items.
17 . The memory of claim 16 , comprising incorporating adaptive pairwise preference feedback into the latent factor model, such that previous feedback on the pairs of different content items affects future pairs of different content items.
18 . The memory of claim 17 , wherein the latent factor model operates in a Bayesian framework.
19 . The memory of claim 16 , wherein the operations comprise generating a recommendation for content based on the successive responses.
20 . The memory of claim 16 , wherein the operations comprise querying for one question in the sequence of questions.
21 . The memory of claim 16 , wherein the operations comprise predicting a difference in the preference for the pair of different content items.