Back to results
US 11,080,596 B1Granted
Prediction filtering using intermediate model representations
AssigneeAmazon Technologies, Inc.
InventorsRoshan Harish Makhijani, Soo-Min Pantel, Sanjeev Jain, Gaurav Chanda
Life of this patent
Estimating the term…
Abstract
The present disclosure is directed to filtering co-occurrence data. In one embodiment, a machine learning model can be trained. An output of an intermediate structure of the machine learning model (e.g., an output of an internal layer of a neural network) can be used as a representation of an event. Similarities between representations of events can be determined and used to generate, augment, or modify co-occurrence data.
Claims20
A system comprising:
- computer-readable memory storing executable instructions;
- and
- one or more hardware-based processors programmed by the executable instructions to at least:
- obtain behavior data representing a first type of behavior of a first user with respect to items of a plurality of items;
- generate first training data for an embedding space artificial neural network using the behavior data, wherein the first training data comprises a first training input and a first corresponding truth data, wherein an element of the first training input corresponds to data representing the first type of behavior of the first user with respect to a first item of the plurality of items at a first time, wherein an element of the first truth data corresponds to data representing the first type of behavior of the first user with respect to a second item of the plurality of items at a second time, the second time occurring within a predetermined period of time after the first time;
- generate a training output using the embedding space artificial neural network and the training input;
- compute a difference between the training output and the truth data;
- update a parameter of an internal layer of the embedding space artificial neural network using the difference between the training output and the truth data;
- obtain second training data for a prediction artificial neural network, wherein the second training data comprises data representing a likelihood of co-occurrence of the first type of behavior of a second user with respect to both the first item and the second item;
- determine a first intermediate vector comprising a first representation of the internal layer of the embedding space artificial neural network using a first input, wherein an element of the first input corresponds to data representing the first item;
- determine a second intermediate vector comprising a second representation of the internal layer of the embedding space artificial neural network using a second input, wherein an element of the second input corresponds to data representing the second item;
- determine that a distance between the first intermediate vector and the second intermediate vector satisfies a threshold;
- generate a modified likelihood of co-occurrence using:
- the distance between the first intermediate vector and the second intermediate vector,
- the data representing the likelihood of co-occurrence of the first type of behavior of the second user with respect to both the first item and the second item, and
- one or more multiplication factors, wherein the one or more multiplication factors are applied to adjust at least one of the distance or the data representing the likelihood of co-occurrence;
- and
- train the prediction artificial neural network using the second training data and the modified likelihood of co-occurrence.
15 more elements
A computer-implemented method comprising:
- under control of a computer system comprising one or more hardware processors configured to execute specific computer-executable instructions,
- obtaining first training data for an embedding space machine learning model using data regarding one or more occurrences of a first type of event with respect to individual items of a plurality of items;
- training the embedding space machine learning model using the first training data;
- obtaining co-occurrence data comprising first data representing a first likelihood of co-occurrence of the first type of event with respect to a first item of the plurality of items and the first type of event with respect to a second item of the plurality of items;
- determining a first intermediate vector representing a first output of an intermediate structure of the embedding space machine learning model using a first input comprising an element that corresponds to data representing the first item;
- determining a second intermediate vector representing a second output of an intermediate structure of the embedding space machine learning model using a second input comprising an element that corresponds to data representing the second item;
- determining a distance between the first intermediate vector and the second intermediate vector;
- and
- generating second data representing a second likelihood of co-occurrence of the first type of event with respect to the first item and the second item, wherein the second data is generated using:
- the distance,
- the first likelihood of co-occurrence, and
- one or more multiplication factors, wherein the one or more multiplication factors are applied to adjust at least one of the distance or the first likelihood of co-occurrence.
9 more elements
A system comprising:
- computer-readable memory storing executable instructions;
- and
- one or more processors programmed by the executable instructions to at least:
- obtain a machine learning model, wherein the machine learning model is trained, using input data associated with an occurrence of a first type of event with respect to a first object, to generate output data representing a predicted occurrence of the first type of event with respect to a second object;
- obtain first co-occurrence data representing a first likelihood of co-occurrence of the first type of event with respect to the first object and the first type of event with respect to the second object;
- determine a first embedding, representing the first type of event with respect to the first object, and a second embedding, representing the first type of event with respect to the second object, using the machine learning model;
- determine a distance between the first embedding and the second embedding;
- and
- generate second co-occurrence data representing a second likelihood of co-occurrence of the first type of event with respect to the first object and the first type of event with respect to the second object, wherein the second co-occurrence data is generated using:
- the distance,
- the first likelihood of co-occurrence, and
- one or more multiplication factors, wherein the one or more multiplication factors are applied to adjust at least one of the distance or the first likelihood of co-occurrence.
9 more elements
Dependencies are inferred from the claim text.