Machine-learned desking vehicle recommendation
Systems and methods are disclosed herein for machine-learned vehicle desking operations. A vehicle recommendation system receives a request to determine similarities between vehicles. The request can indicate an identifier of a user-specified vehicle associated with vehicle attribute values (e.g., white color, sedan body style, 2020 manufacturing year, etc.). A machine learning model can determine respective embeddings for the vehicle attribute values and the respective embeddings can be concatenated, where the concatenated embeddings represent the user-specified vehicle in one embedding. The system can determine similarity metrics of the concatenated embeddings against reference embeddings. For example, a cosine similarity value can be determined for the concatenated embedding of the user-specified vehicle and the respective reference embeddings. Each similarity metric can represent a measure of similarity between the user-specified vehicle and a given vehicle. The vehicle recommendation system provides for display identifiers of vehicles that are ranked based on the determined similarity metrics.
1. A non-transitory computer-readable medium comprising memory with instructions encoded thereon, the instructions, when executed, causing one or more processors to perform operations, the instructions comprising instructions to:
receive a request to determine similarities between candidate vehicles and a user-specified vehicle associated with a plurality of vehicle attributes corresponding to a plurality of vehicle attribute values;
calculate a conditional probability between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute;
select a subset of the plurality of vehicle attribute values using the conditional probability;
determine, using a machine learning model and the subset of the plurality of vehicle attribute values, a plurality of similarity metrics representative of similarities between the user-specified vehicle and the candidate vehicles; and
provide for display identifiers of one or more candidate vehicles ranked based on the plurality of similarity metrics.
2. The non-transitory computer-readable medium of claim 1 , wherein the machine learning model is trained to receive an input vector of a first dimension representing training vehicle attribute values and output an embedding vector of a second dimension representing the training vehicle attribute values, the second dimension smaller than the first dimension.
3. The non-transitory computer-readable medium of claim 1 , wherein the plurality of vehicle attribute values are a first plurality of vehicle attribute values, and wherein the instructions further comprise instructions to:
generate, using the machine learning model, embeddings using a second plurality of vehicle attribute values from an inventory of vehicles; and
store the generated embeddings, wherein a plurality of reference embeddings include the generated embeddings, the plurality of reference embeddings associated with the candidate vehicles.
4. The non-transitory computer-readable medium of claim 1 , wherein the plurality of vehicle attribute values includes one or more of a color, a make, a model, a year, or a safety feature.
5. The non-transitory computer-readable medium of claim 1 , wherein the instructions further comprising instructions to determine, using the conditional probability, mutual information between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute.
6. The non-transitory computer-readable medium of claim 5 , wherein the instructions to select the subset of the plurality of vehicle attribute values using the conditional probability comprise instructions to select for input to the machine learning model, for each of a predefined number of the plurality of vehicle attributes of having high mutual information, a vehicle attribute value corresponding to the vehicle attribute.
7. The non-transitory computer-readable medium of claim 1 , wherein the machine learning model is a neural network comprising a plurality of hidden layers, each hidden layer comprising a plurality of hidden nodes, wherein the instructions further comprise instructions to train the neural network by determining weights associated with connections between the plurality of hidden nodes to minimize a loss function such that generated embeddings are representative of similar valuation amount ranges for attributes mapped to an embedding.
8. The non-transitory computer-readable medium of claim 7 , wherein the instructions further comprise instructions to re-train the neural network by adjusting a dimension of a hidden layer of the plurality of hidden layers to minimize a loss function of the neural network.
9. The non-transitory computer-readable medium of claim 7 , wherein the instructions further comprise instructions to:
receive user feedback; and
re-train the neural network by adjusting the weights based on the received user feedback.
10. The non-transitory computer-readable medium of claim 7 , wherein the instructions further comprise instructions to determine that the loss function is minimized by determining that a mean additive error between a valuation amount of a recommended vehicle output by the neural network and a valuation amount of a user-specified vehicle is below a threshold error.
11. The non-transitory computer-readable medium of claim 1 , wherein the instructions further comprise instructions to re-train the machine learning model responsive to an update of an automotive inventory database in which a plurality of vehicle attribute-value pairs are stored.
12. The non-transitory computer-readable medium of claim 1 , wherein the instructions to determine, using the machine learning model and the subset of the plurality of vehicle attribute values, further comprise instructions to calculate respective cosine similarities between concatenated embeddings output by the machine learning model and a plurality of reference embeddings.
13. A method comprising:
receiving a request to determine similarities between candidate vehicles and a user-specified vehicle associated with a plurality of vehicle attributes corresponding to a plurality of vehicle attribute values;
calculating a conditional probability between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute;
selecting a subset of the plurality of vehicle attribute values using the conditional probability;
determining, using a machine learning model and the subset of the plurality of vehicle attribute values, a plurality of similarity metrics representative of similarities between the user-specified vehicle and the candidate vehicles; and
providing for display identifiers of one or more candidate vehicles ranked based on the plurality of similarity metrics.
14. The method of claim 13 , wherein the machine learning model is trained to receive an input vector of a first dimension representing training vehicle attribute values and output an embedding vector of a second dimension representing the training vehicle attribute values, the second dimension smaller than the first dimension.
15. The method of claim 13 , further comprising determining, using the conditional probability, mutual information between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute.
16. The method of claim 15 , wherein selecting the subset of the plurality of vehicle attribute values using the conditional probability comprises:
selecting for input to the machine learning model, for each of a predefined number of the plurality of vehicle attributes of having high mutual information, a vehicle attribute value corresponding to the vehicle attribute; and
adjusting the predefined number to minimize a loss function of the machine learning model.
17. A system comprising:
memory with instructions encoded thereon; and
one or more processors that, when executing the instructions, are caused to perform operations comprising:
receiving a request to determine similarities between candidate vehicles and a user-specified vehicle associated with a plurality of vehicle attributes corresponding to a plurality of vehicle attribute values;
calculating a conditional probability between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute;
selecting a subset of the plurality of vehicle attribute values using the conditional probability;
determining, using a machine learning model and the subset of the plurality of vehicle attribute values, a plurality of similarity metrics representative of similarities between the user-specified vehicle and the candidate vehicles; and
providing for display identifiers of one or more candidate vehicles ranked based on the plurality of similarity metrics.
18. The system of claim 17 , wherein the machine learning model is trained to receive an input vector of a first dimension representing training vehicle attribute values and output an embedding vector of a second dimension representing the training vehicle attribute values, the second dimension smaller than the first dimension.
19. The system of claim 17 , wherein the operations further comprise determining, using the conditional probability, mutual information between a vehicle attribute of the plurality of vehicle attributes and a valuation amount of the vehicle attribute.
20. The system of claim 19 , wherein selecting the subset of the plurality of vehicle attribute values using the conditional probability comprises:
selecting for input to the machine learning model, for each of a predefined number of the plurality of vehicle attributes of having high mutual information, a vehicle attribute value corresponding to the vehicle attribute; and
adjusting the predefined number to minimize a loss function of the machine learning model.