Systems and methods for preference and similarity learning
Systems and methods for preference and similarity learning arc disclosed. The systems and methods improve efficiency for both searching datasets and embedding objects within the datasets. The systems and methods for preference embedding include identifying paired comparisons closest to a user's true preference point. The processes include removing obvious paired comparisons and/or ambiguous paired comparisons from subsequent queries The systems and methods for similarity learning include providing larger rank orderings of tuples to increase the context of the information in a dataset In each embodiment, the systems and methods can embed user responses in a Euclidean space such that distances between objects are indicative of user preference or similarity.
1 . A method of embedding a user preference comprising:
receiving, at an input/output interface, a dataset comprising a plurality of items;
embedding, with a processor, the plurality of items as coordinate vectors in a d-dimensional Euclidean space;
generating, with the processor, a similarity matrix, the similarity matrix comprising Euclidean distances between coordinate vectors of the plurality of items;
actively selecting informative queries by:
identifying, with the processor, at least two paired comparisons of a first item and a second item of the plurality of items for transmittal to a user device for comparison;
receiving, at the input/output interface, for each of the at least two paired comparisons, a user preference of the first item over the second item to create a user preference point; and
subsequent to receiving the at least two paired comparisons, Euclidean embedding, with the processor, each user preference point in the d-dimensional Euclidean space with respect to the coordinate vectors such that distances between each user preference point and respective items of the plurality of items are indicative of corresponding user preference; and
identifying a subsequent paired comparison by using the similarity matrix generated from the embeddings of the plurality of items in combination with distances between user preference points and corresponding items of the plurality of items in the d- dimensional Euclidean space.
2 . The method of claim 1 , wherein the dataset is received from a training set of triplet comparison queries.
3 . The method of claim 2 , wherein the training set of triplet comparison queries is indicative of a Crowd Kernel Learning technique or a Stochastic Triplet Embedding technique.
4 . The method of claim 1 , wherein the dataset is received from a plurality of tuple comparisons.
5 . The method of claim 1 , wherein identifying the paired comparison comprises ignoring, with the processor, ambiguous paired comparisons in the plurality of items.
6 . The method of claim 1 , wherein identifying the paired comparison is based on an equiprobable, max-variance scheme.
7 . The method of claim 1 , wherein identifying the paired comparison is based on a mean cut, max-variance scheme.
8 . The method of claim 1 further comprising identifying, with the processor, a third item that corresponds to the user preference, the identification based on a proximity of the third item to the user preference in the space.
9 . The method of claim 8 , wherein identifying, with the processor, the paired comparison minimizes a quantity of queries required to identify the third item.