Generating and utilizing classification and query-specific models to generate digital responses to queries from client device
The present description relates to systems, methods, and non-transitory computer readable media for generating digital responses to digital queries by utilizing a classification model and query-specific analysis models. For example, the described systems can train a classification model to generate query classifications corresponding to product queries, conversational queries, and/or recommendation/purchase queries. Moreover, the described systems can apply the classification model to select pertinent models for particular queries. For example, upon classifying a product query, the described systems can utilize a neural ranking model (trained based on a set of training product specifications and training queries) to generate relevance scores for product specifications associated with a digital query. The described systems can further compare generated relevance scores to select a product specification and generate a digital response that includes the pertinent product specification to provide for display to a client device.
1. A system comprising:
at least one processor; and
a non-transitory computer readable medium comprising instructions that, when executed by the at least one processor, cause the system to:
identify a training query that corresponds to a first product specification of a training product, wherein the training product corresponds to a plurality of training product specifications;
train a neural ranking model comprising a bidirectional long short-term memory layer and a differential product concatenation layer to generate relevance scores between product specifications and digital queries by, for each training product specification of the plurality of training product specifications:
generating a training concatenated vector utilizing the differential product concatenation layer to concatenate a difference vector representing a difference between a training query vector and a training product specification vector together with a product vector representing a product of the training query vector and the training product specification vector;
generating a relevance score indicating a measure of correspondence between the training product specification and the training query by utilizing the bidirectional long short-term memory layer to process the training concatenated vector; and
comparing the relevance score to a ground truth score for the training product specification, the ground truth score indicating whether the training product specification is the first product specification that corresponds to the training query.
2. The system of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the system to train the neural ranking model by further generating, utilizing a word representation layer of the neural ranking model:
a training query text vector from the training query; and
for each of the plurality of training product specifications, a training specification text vector from the training product specification.
3. The system of claim 2 , further comprising instructions that, when executed by the at least one processor, cause the system to train the neural ranking model by further generating, utilizing the bidirectional long short-term memory layer:
a training query vector from the training query text vector and representing the training query; and
for each of the plurality of training product specifications, a training product specification vector from the training specification text vector and representing the training product specification.
4. The system of claim 3 , further comprising instructions that, when executed by the at least one processor, cause the system to train the neural ranking model by further utilizing the differential product concatenation layer to generate, for each of the plurality of training product specifications:
the difference vector by determining a difference between the training query vector and the training product specification vector; and
the product vector by determining a product of the training query vector and the training product specification vector.
5. The system of claim 1 , further comprising instructions that, when executed by the at least one processor, cause the system to train the neural ranking model by further utilizing a full connection layer and a softmax layer of the neural ranking model to determine, for each of the plurality of training product specifications, a probability that the training product specification is relevant to the training query.
6. The system of claim 1 , wherein comparing the relevance score to the ground truth score comprises applying a loss function to determine a measure of loss between the relevance score and the ground truth score.
7. The system of claim 6 , further comprising instructions that, when executed by the at least one processor, cause the system to train the neural ranking by modifying parameters of the long short-term memory layer and a full connection layer to reduce the measure of loss between the relevance score and the ground truth score.