Predicting labels using a deep-learning model

One or more computer-readable non-transitory storage media embodying software that is operable when executed to:

• receive, from a client system, a text input comprising a plurality of n-grams;
• accessing a deep-learning model, wherein the deep-learning model is trained to map n-grams to respective vector representations using a supervised machine learning algorithm, wherein the supervised machine learning algorithm has a loss function determined by a difference of a similarity score for a plurality of positive sample labels and a similarity score for a plurality of negative sample labels, and wherein the plurality of positive sample labels are labels known to be relevant to the n-grams and the plurality of negative sample labels are randomly selected labels;
• determining, using the deep-learning model, a vector representation of the text input based on respective vector representations of the plurality of n-grams, wherein each of the plurality of n-grams is mapped to one or more vector representations, respectively, using a first dictionary generated by the deep-learning model;
• determining, using the deep-learning model, an embedding of the vector representation of the text input in a d-dimensional embedding space based on a combination of the one or more vector representations of the each of the plurality of n-grams of the text input using a max pooling operation;
• identify one or more labels based on, for each of the one or more labels, a respective similarity of an embedding of a vector representation of the label in the embedding space to the embedding of the vector representation of the text input;
• and send, to the client system in response to the received text input, instructions for presenting a user interface comprising one or more of the identified labels as suggested labels.

A system comprising:

• one or more processors;
• and
• a memory coupled to the processors comprising instructions executable by the processors, the processors operable when executing the instructions to:
• receive, from a client system, a text input comprising a plurality of n-grams;
• accessing a deep-learning model, wherein the deep-learning model is trained to map n-grams to respective vector representations using a supervised machine learning algorithm, wherein the supervised machine learning algorithm has a loss function determined by a difference of a similarity score for a plurality of positive sample labels and a similarity score for a plurality of negative sample labels, and wherein the plurality of positive sample labels are labels known to be relevant to the n-grams and the plurality of negative sample labels are randomly selected labels;
• determine, using the deep-learning model, a vector representation of the text input based on respective vector representations of the plurality of n-grams, wherein each of the plurality of n-grams is mapped to one or more vector representations, respectively, using a first dictionary generated by the deep-learning model;
• determine, using the deep-learning model, an embedding of the vector representation of the text input in a d-dimensional embedding space based on a combination of the one or more vector representations of the each of the plurality of n-grams of the text input using a max pooling operation;
• identify one or more labels based on, for each of the one or more labels, a respective similarity of an embedding of a vector representation of the label in the embedding space to the embedding of the vector representation of the text input;
• and send, to the client system in response to the received text input, instructions for presenting a user interface comprising one or more of the identified labels as suggested labels.